Team:Grenoble Alpes/Results

PyoBuster - Results
Constitutive + Therapeutic molecules Plsr + E7 lysis Constitutive + RhlR Prhlr + LacI Genomic DNA Ptac + E7 lysis
Constitutive + Therapeutic molecules Plsr + E7 lysis Constitutive + RhlR Prhlr + LacI Genomic DNA Ptac + E7 lysis

Wet Lab Results

PyoBusters is a genetically engineered E. coli strain, fulfilling a therapeutic action against P. aeruginosa and its biofilm. Our PyoBusters system works thanks to a plasmid comprising 4 different sets of genes, as well as a gene inserted in the genome.

Those sets of genes can be separated into 3 systems, operating in parallel with different objectives.

  • Survival system: Specifically target the P. aeruginosa’s biofilm to grow on its surface.
  • Therapeutic system: Produce intracellular therapeutic molecules to destroy P. aeruginosa and its biofilm.
  • Delivery system: Work as a delivery system allowing the in-situ release of therapeutic molecules.

WetLab Introduction

The main objective of our experiments was to design a PyoBusters able to detect Pseudomonas aeruginosa’s biofilm in order to trigger gene expression. We intended, besides, to produce two therapeutic molecules capable of destroying the biofilm and killing P. aeruginosa. We also wanted to show that it is possible to use E. coli's sensing quorum to induce the PyoBusters lysis thus enabling the release of the therapeutic molecules.

Survival system

1. Constitutive promoter for the production of the RhlR

In order to compare the strength of three promoters, we have monitored four DH5alpha cultures:

These 4 cultures were inoculated in triplicata (OD600nm=0.1) in a semi-transparent 96-wells plate and incubated at 37°C in Varioskan LUX six hours long. This experiment was repeated three times and data were collected and analysed together to build the following graphs.

Every ten minutes, both OD600nm (Figure 1) and eGFP fluorescence (Figure 2) were measured in each well and triplicata's mean were plotted.

evolution of DH5a OD overtime
Figure 1: Optical density of DH5α over time according to the J23 promoter

Growth curves are presented Figure 1. We can see that all these curves have the same pattern of evolution over time. Therefore, when compared to the growth curve of the control culture, we can conclude that the strength of a promoter has no effect on the growth of the bacterial culture (p-value=0.896).

fluo J23
Figure 2: kinetics of eGFP expression under the control of various promoters

Averages: J23100>J23102>J23106>DH5wt

The figure 2 shows the fluorescence intensity over time. For each curve, the first measurement point (T0) was set at 0 and subtracted to the following ones. Firstly, no fluorescence was detected in the DH5 alpha wt control culture. In contrast, in all the cultures expressing eGFP under the control of a specific constitutive promoter, we detected an significant increasing amount of fluorescence over time (p-value<2.10-16). The total amount of eGFP produced increased according to the promoter tested with 15.2, 9.7 and 2.5 RFU for J23100, J23102 and J23106 respectively. Significantly, the promoter J23100 seems to be the best one to produce eGFP in large quantities. Thus, it will be used to produce RhlR BBa_K3463007.

2. Inducible promoter rhlR (PrhlR)

The Figure 3 illustrates the fluorescence intensity over time for 2 growing cultures of E. coli Nissle transformed with the plasmid pUCBB PrhlR-eGFP BBa_K3463011. Cultures were performed in absence or in presence of 100µM of synthetic BHL. In addition, a culture of wild type E. coli Nissle (not transformed) was used as a negative control. The fluorescence intensity was monitored for 15 hours. Each time point was performed hourly in triplicata. For each curve, the first measurement (T0) was set at 0 and subtracted to following ones.
This experiment was repeated 3 times.

anova results
Figure 3: eGFP expression according to the BHL and over time. Two conditions with 100µM of BHL and without BHL were performed to evaluate the effect of BHL in E. coli Nissle. With and without BHL, no eGFP expression should be detected because there is no production of RhlR protein.

Averages: Without BHL=100µM BHL=Control -

As it can be seen in figure 3, all cultures have similar levels of eGFP expression over time (p-value=0,44). Indeed, we can see that the fluorescence intensity of E. coli Nissle PrhlR-eGFP increases slightly whatever the presence of BHL in the medium. In addition, the curve of negative control also increases in the same manner over time. Therefore, there is no significant expression of eGFP in E. coli Nissle PrhlR-eGFP, even with a large amount of BHL in the medium.

To conclude, we can significantly ensure that our PrhlR construction BBa_K3463011 does not allow E. coli Nissle to express downstream genes without RhlR, whatever the presence of BHL.

From this step, the proof of concept’s experiment using pUCBB J23100-RhlR-B0015-PrhlR-eGFP can be done.

3. BHL sensing

To evaluate if our system works, we performed 6 bacterial cultures of transformed E. coli Nissle BBa_K3463019 incubated with increasing amounts of synthetic BHL in the medium from 10nM to 100µM. Once all the media have been set up at T0 with 100µM / 10µM / 1µM / 100nM / 10nM and without BHL, bacteria were inoculated at OD600nm=0.1.Then, fluorescence measurements of eGFP were performed for 15H (hourly in triplicata until 7H) in opaque 96-wells (FluoStars). For each curve in figure 4, the first measurement point (T0) was set at 0 and subtracted to the following ones. This experiment was repeated 3 times.

In addition, a culture of wild type E. coli Nissle (not transformed) was used as a negative control.

BHL fluo
Figure 4: eGFP expression according to the BHL concentration and over time. Different concentrations of BHL were performed to evaluate the effect of BHL in E. coli Nissle. Thanks to the BHL, the engineered E. coli should be able to express eGFP. Without BHL, no eGFP expression should be detected.

Averages: 100µM/10µM/1µM > others

The Figure 4 shows the fluorescence intensity of each bacterial culture according to the BHL concentration and over time. While the wild type E. coli Nissle auto-fluorescence is used as blank, all the other transformed cultures show significant eGFP expression.

First, we can see that transformed E. coli Nissle produce low quantities of eGFP even without BHL. However, we can also observe that the more BHL, the more eGFP expression (pvalue<2.10-16). Indeed, in the culture containing 10µM or 100µM of BHL, the fluorescent intensity starts to increase sharply after 3 hours of culture and reaches 800 RFU or more. Nevertheless, even if a curve reaches 600 RFU with 1µM, weaker concentrations of BHL (100nM) don't allow transformed E. coli Nissle to significantly express eGFP.

Through an ANOVA test, we can conclude that, thanks to the new part BBa_K3463019, our engineered E. coli Nissle is able to sense low concentration of BHL (1µM) in its environment and express a specific gene under the control of PrhlR.

4. Ptac construction and verification

To validate the UTR designer prediction and check the B0017 relevance, we measured the fluorescence and the absorbance of the cultures of the strains containing one of the following constructs: PtacHigh-GFP, PtacLow-GFP-B0017 BBa_K3463018, PtacLow-GFP-B0017 BBa_K3463017, and a non-transformed strain as a negative control for to 4,5h. The experiment was carried out 3 times.

We obtained the following result shown in figure 5.

survival ptac
Figure 5: Effect of UTR prediction and B0017 terminator on eGFP fluorescence expression. The Figure 5 shows that no fluorescence was detected in the culture of the control non- transformed strain. As predicted by UTR designer the strong 5’UTR sequence (high) gave higher fluorescent levels than the weak 5’UTR sequence (Low). The terminator B0017 seems to have a limited effect on the fluorescent levels. The last seems to only impact the eGFP production of the strong UTR sequence.

First, the influence of the UTR over the fluorescent level obtained was confirmed by an Anova displayed in figure 6.

anova result
Figure 6: Anova results of constructs influence over fluorescent. Constructs have significant influence over fluorescent: P-value <2.10-16.

And secondly, as predicted in-silico by UTR designer, the HIGH_5’UTR has a significantly stronger expression level than the LOW_5’UTR. However, during our experiment, we didn’t find any significant impact of the B0017 terminator on the fluorescence.
It should be noted that the relative fluorescence level between the difference constructions seems to be slightly different from our prediction with a HIGH relative expression of 100% and a LOW relative of 50%. According to our model, the relative expression of the LOW prediction was around 75% for a HIGH fixed at 100%. A part of this difference can be explained by the prediction incertitude. But further investigation and experiments need to be done in order to fully assess the origin of this difference.
This experiment opens new ways to future iGEM participants that could use a reliable tool to predict protein expression and therefore give them the ability to build complex genetic networks.

5. Genomic integration of the constructions

The genomic integration was made following the Datsenko et al. protocol adjusted to fit our needs. Initially, the integration protocol was carried out without using the resistance gene to limit the use of antibiotics and avoid the creation of strain carrying a resistance gene into its genome. This approach was proved to be ineffective as it gave too many colonies to screen. Indeed a 1/10-6 dilution was necessary to obtain screenable colonies on LB-Agar. No colonies showed visible green fluorescence when illuminated at 475nm. The experiment was carried out three times and gave the same data. Consequently, the protocol was carried out following the initial protocol Datsenko. The PKD3 chloramphenicol gene resistance flanked by the 2 FRT sites, was amplified by PCR and added to the Plasmid pUCBB PtacHigh-GFP-B0017 or PtacLow-GFP-B0017 between the restriction sites XhoI and SpeI. The overall construct that needed to be integrated into the genome namely pUCBB PtacHigh/Low-eGFP-B0017 was then unsuccessfully amplified by PCR. This failure was identified as a primer issue, caused by a possible mutation or recombination happening during the insertion of the PKD3 resistance gene into the pUCBB PtacHigh-eGFP-B0017 or PtacLow-eGFP-B0017.

survival genomic integration
Figure 7: PCR amplification of the genomic insert. 1: PtacHigh-eGFP-B0017-PKD3 2: control PCR 3: PtacHigh-eGFP-B0017.

This mutation was highlighted by PCR. The genomic insert was able to be produced only before the addition of PKD3 gene resistance as seen in figure 7.
Knowing that the amplified zone is set a few nucleotides before the EcoRI restriction site in 5’ and after the SpeI one in 3’ with the primer beginning at the last nucleotides of the SpeI site, we assumed that this SpeI site was involved in the failure.
This specific SpeI site was often mutated during our cloning (cf 6. Proof of concept construction), which suggests that this restriction site is responsible for the PCR failure.
We should have designed the insert primer out of the restriction sites to avoid any mutation impacting the PCR.
Further investigation needs to be done to fully understand the encountered problem such as sequencing.

6. Proof of concept construction

To get a proof of concept for our all genetic circuit, we decided to build a Proof of concept plasmid BBa_K3463020 containing all the parts of the survival system:

We then transformed a BL21 strain with BBa_K3463020 and monitored the expression of the eGFP by measuring the level of fluorescence under a gradient of concentration of synthetic BHL: 100µM / 10µM / 1µM / 100nM and without BHL.

res survival poc
Figure 8: Effect of the BHL on the Proof of concept construct. No significant differences were obtained between the tested concentration and the negative control.

In figure 8, the BHL didn’t have any effects on the fluorescent level (p-value=0.999). Different hypotheses can be made to explain these data. For instance the genetic networks didn’t function as planned. However different explanations can be given before coming to this conclusion. First the fluorescent expression level obtained is much higher than those obtained classically with the Ptac-GFP construct and much in line with the level obtained with the classic pUCBB-GFP suggesting that the construct doesn’t contain the tac promoter. However the plasmid was built following the standard assembly protocol and gave the expected results for the digestion gel control (figure 9 and 10).
Despite having the desired digestion profile some screened colonies show an usual profile with missing restriction sites. To be more precise the PstI and SpeI restrictions were missing as shown in figures 9 and 10. That loss of restriction sites can be due to a potential rearrangement or alternative cut made by the restriction enzyme.
Besides, after the second standard assembly the tac promoter couldn’t be amplified by PCR whereas it could be from the first standard assembly step (not shown). This suggested a mutation or a modification of the Ptac during the second standard assembly explaining the fluorescent level and the dysfonctionnement of our genetic network.
A DNA sequencing is necessary to fully understand the issue concerning our proof of concept build, although we had no remaining time to perform it.

res gel survival poc
Figure 9: Digestion of the positive colonies pUCBB Ptac-eGFP-B0017-J23100-RhlR-B0015 in EcoRI and PstI. All colonies were negative with colonies 3 and 5 as classic pUCBB but 1-2 and 5 showed a missing PstI restriction sites after the standard assembly.
res gel survival 2 poc
Figure 10: Digestion of the positive colonies pUCBB Ptac-eGFP-B0017-J23100-RhlR-B0015-Prhlr-LacI BBa_K3463020 in EcoRI and PstI. The 1-2-3-6-8-9 colonies were negative i.e. classic pUCBB. The 5 was positive and contained the 3 composite parts. The 4-7 showed an unexpected profile with a missing PstI restriction site.

Therapeutic system

1. Amplification of the target genes and construction

To do the construction with the Dispersin B, we used the part BBa_K1659200, designed by Wei Chung Kong iGEM15_Oxford15 (2015-08-28).
Concerning the constructions containing the Alginate lyase (BBa_K3463002) and the Pyocin S5 (BBa_K3463001) genes, we extracted them from the PAO1’s genome. To do so, we used a kit for genome extraction (see protocols) then proceeded to PCR amplification with appropriate primers. As you can see (figure 11), the extraction of the Alginate lyase shows non-specific bands but we extracted the band at the right size (1326 bp). We extracted the Pyocin S5 at the right size (1512 bp) too.

res therapie gel
Figure 11: PCR amplification of the Alginate lyase and Pyocin S5 from the genomic extraction material. 1: control PCR Alginate lyase, 2 and 3: Alginate lyase, 4: control PCR Pyocin S5, 5 and 6: Pyocin S5. The amplification of these genes was conclusif.

We wanted to clone the sequence of our three therapeutic molecules in both pUCBB-J23100-eGFP and pUCBB-ntH6-eGFP.
As we said in the Experiment section, we used primers with particular restriction sites to standardize the cloning strategy.
Unfortunately we didn’t succeed to clone the Alginate lyase gene in the pUCBB-ntH6-eGFP. It may be explained by the PCR’s product that was not pure enough. Indeed, as you can see on figure 12, several non-specific bands are still present in the insert sample. It can be explained by the fact that the Tm or the primers were not optimal.

res therapie gel 2
Figure 12: PCR amplification of the Alginate lyase. 1: Alginate lyase, 2: control PCR Alginate lyase. The amplification shows several non-specific bands.

The Dispersin B and the Pyocin S5 were respectively cloned in pUCBB-J23100-eGFP and pUCBB-ntH6-eGFP in E. coli BL21 to optimize protein expression successfully.

2. Tests on PAO1

As we didn’t manage to create the PAO1’s biofilm in our laboratory, we sent the Dispersin B BBa_K1659200 to evaluate its activity on P. aeruginosa’s biofilm to Bioaster.
For that, their tested conditions were:

  • Lysis buffer (used to obtain the E. coli lysates)
  • Lysate Dispersin B
  • Lysate Pyocin S5
  • Lysate non-transformed E. coli BL21

These conditions were both inoculated at 25% (163µL of component + 487µL of Dulbecco's phosphate-buffered saline) or 50% (325µL of component + 325µL of DPBS) in triplicate and compared to a non-treated condition.
Then, they counted the CFU/mL per replicate and calculated the bacterial concentrations for each condition.
As you can see (figure 13), the preliminary data are highly dispersed (cf standard deviations) for the tested conditions. The lysis buffer condition didn’t impact bacterial concentrations and comparable results to the non-treated conditions. Concerning the Dispersin B and the Pyocin S5, they present a dose-effect and the bacterial concentrations are higher than the non-treated condition. It seems that they have a stimulatory effect.

therapeutic molecules effects
Figure 13: Effect of the therapeutic molecules on the Pseudomonas aeruginosa’s biofilm resuspension in phosphate-buffered saline. The CFU/mL per replicate were counted after resuspension of the P. aeruginosa biofilm. The bacterial concentrations were calculated for each condition.

The next step would be to reiterate these experiments by optimizing the conditions. To do so, the lysates would be purified to obtain higher concentrations of the therapeutic molecules than in the preliminary tests and optimize the sonication protocol to release the molecules. We could do a western blot with antibodies directed against both Dispersin B and Pyocin S5 to make sure that the proteins are expressed.
Moreover, the freezing/thaw process for the transport can have an impact on the protein and have to be avoided.
Furthermore, in our experimental condition, the observed effect of the three therapeutic molecules didn’t correlate with the one observed in the literature. Therefore further experiments could allow us to fully understand and observe the mechanism of those molecules and to assess their relevance in our therapeutic system.

therapetics tubes
Figure 14: Killing effect of the Pyocin S5 on planktonic Pseudomonas aeruginosa.

To evaluate the killing activity of the Pyocin S5 BBa_K3463001 on P. aeruginosa, we performed two different experiments:
The first one (Figure 14) aimed to show the OD decrease of P. aeruginosa’s growing cultures over time incubated with various concentrations of Pyocin S5 and 3 negative controls.

The data obtained from this experiment (Figure 15) didn’t show any differences between all the growing cultures after 2H of incubation. Indeed, the OD of P. aeruginosa didn’t decrease or decelerate over the time for cultures with Pyocin S5. Thus, we can say that the Pyocin S5 doesn’t have any killing activity against planktonic P. aeruginosa in suspension at the concentration tested.

therapetics table
Figure 15: Table representing the growth rate of pseudomonas aeruginosa under six different conditions.

The second experiment (Figure 16) showed the same result on plate. We wanted to show if the Pyocin S5 can block the growth of P. aeruginosa on LB agar plates.

After an overnight culture at 37°C, the plate was analysed and didn’t show any growing inhibition around all the 5 discs soaked with Pyocin S5. Thus, we can say that the Pyocin S5 doesn’t have any killing or inhibiting activity against P. aeruginosa on LB agar plate.
Nevertheless, these unexpected results can be explained by the sonication protocol (not optimised) and the stockage at -80°C. Indeed, it is possible that these steps led to a loss of the activity of the Pyocin S5 protein.

therapeutics petri test
Figure 16: Agar plate representing the killing effect of five different samples on Pseudomonas aeruginosa.

Delivery system

1. UTRs comparaison

To confirm the influence of the optical density on the lsrA promoter, we compared the fluorescent expression of mCherry due to a constitutive J23100 and PlsrA.

For this experiment all the construction PUCBB-J23100-mCherry, pUCBB-PlsrHigh-mCherry (BBa_K3463023 ) used were successfully obtained and transformed in E. coli BL21.

Each bacteria was cultivated at 0,5 OD in LB and we measured the optical density and the mCherry fluorescence hourly with the FluoStar BMG labtech (we made triplicate measurement ). An Opaque-walled 96 was used and the fluorescent gain had been set to 2000 (we made triplicate measurement). This experience was repeated three times in the same conditions.

rdelievry lsr promoteurA delievry lsr promoteurB
Figure 17: mCherry fluorescence expression reported at bacterial optical density over time. To represent the two curves, we used two gradations: the one on the left represents the expression of the lsr promoter and the one on the right represents the J23100 promoter. A and B are two examples of the data obtained in the same working conditions

Figure number 17 shows the expression of the mCherry under PlsrA, J23100 promoter and BL21 WT as control over time.

We can first observe that the expression of mCherry under the control of PlsrA occurs between 7 and 8 hours after inoculation, while the constitutive promoter expresses mCherry earlier (after only 3 to 4 hours). In addition, the level of expression of mCherry is about 10 times lower under the control of PlsrA rather than under the control of the constitutive promoter J23100 which informs about the strength of the promoter. In addition, the level of expression of mCherry is about 10 times lower if it is under the control of Plsr rather than under the control of the constituent promoter J23100 which is not surprising.

Intending to prove that the Plsr expression is related to Autoinducer 2 (AI-2) we performed the same experiment with E. coli DH5α. This strain is known for not producing any AI-2, and as expected there was no mCherry expression during the growing time (results not shown).

We also calculated the Spearman correlation between fluorescence and optical density. For the constitutive promoter we obtained 0.6263641 and for Plsr 0.4615318. A higher correlation is observed for J23100 which is explained by a constant expression during growth unlike the expression of the lsr promoter.

The delay of expression observed with lsr promoter in BL21 relative to the constitutive promoter J23100 is probably linked to the necessary accumulation of Ai-2 during growth and therefore to the optical density.
Although this experiment was repeated several times under the same conditions we did not obtain the same results as we can notice (figure 17). Even if the trend suggests that the expression of the lsr promoter is linked to the optical density, there are probably other factors that we have involuntarily stimulated that could influence the expression of the protein.

We wanted to verify the expression of the mCherry protein according to the UTR sequence generated with the UTR designer software. Two sequences were generated, one associated with high score (PlsrHigh) and one low score, 50% less than the last (PlsrLow). For this experiment two construction, pUCBB-PlsrHigh-mCherry (BBa_K3463023), pUCBB-PlsrLow-mCherry (BBa_K3463024) used were successfully obtained and transformed in E. coli BL21.
The mCherry expression in E. coli BL21 was compared (figure 18) according to the same protocol as described previously.

delievry fluo
Figure 18: mCherry fluorescence expression under control of PlsrHigh or PlsrLow and WT BL21 reported at bacterial optical density over time.

Figure 18 shows the fluorescent expression associated with the two UTR sequences. We see that the expression of PlsrHigh and PlsrLow is simultaneously triggered around 7.5h. It can also be noted that the difference between the two becomes more and more pronounced.

box plot 1
Figure 19: Boxplot representing the distribution of ttem under the control of Plsr fluorescence as a function of UTR strength using E. coli BL21.

Figure 19 shows a different distribution of mCherry fluorescence values depending on the strength of the UTR sequence. The median for PlsrHigh is higher than that of PlsrLow.
Each time we reproduced this experiment with E. coli BL21 we found a significant difference of fluorescent expression according to the UTR sequence and the 50% ratio was respected.

Once our hypothesis was validated with the E. coli BL21 strain, we wished to verify and see if the difference of expression was preserved in the E. coli Nissle strain.
The same experiment as the previous one, under the same conditions, was reproduced with E. coli Nissle transformed with pUCBB-PlsrHigh-mCherry (BBa_K3463023) and pUCBB-PlsrLow-mCherry (BBa_K3463024).

box plot 1
Figure 20: Boxplot representing the distribution of mCherry under the control of Plsr fluorescence as a function of UTR strength using E. coli Nissle

Figure 20 shows a similar distribution of fluorescence values; using the E. coli Nissle bacterial strain we do not notice a significant difference between mCherry expression with the UTR sequence high or low. E. coli Nissle is not an optimized strain for protein production, unlike BL21 which may explain the lack of expression difference between the two UTR sequences.

All the data obtained suggested that the system is functioning, we have a trigger for the expression of Plsr depending on the E. coli population but it is necessary to control the environment of the culture. Perhaps conducting the culture in bioreactors could permit us to identify other parameters that influence this mechanism. It is important to notice that the fluorescence monitoring was stopped after 10 hours as we could not do more but it would be interesting to see if the difference between high and low becomes more pronounced over time.

2. E. coli lysis

For this part, which concerns the lysis itself, we were able to obtain only one among the ones we had planned to do: construction with Plsr100%-E7 BBa_K3463025.
We succeeded in cloning the lysis gene with the lsrA promoter (results not shown). Unfortunately, we couldn’t optimize the experiment with BBa_K3463025, but we were ready to change strategies.
It is possible to boost the expression of PlsrA by different methods, in particular by expressing the phosphorylated AI-2 degradation proteins as shown by the work of the Diaz Z team [1].

incubator schema

Dry Lab Results

PyoBusters’ goal is to destroy the biofilm of P. aeruginosa which acts in different environments. As engineers we wanted to provide the experimenters a way to grow their biofilms and insert the modified E. coli in an automated experience. Hence we built an incubator called the Automated Measurement Incubator (AMI) which is built and implemented in 4 parts:

  • The environment module that controls and monitor the atmosphere inside the box.
  • The agitation module that operates the plate movements.
  • The fluorescent module that makes the analysis during the whole experiment.
  • The software that controls the 3 previous modules and allows a Human Computer Interface.

DryLab Introduction

The results presented here are associated with the incubator functions: atmosphere controlling and growth bacteria inside our incubator. You will see that our device is able to stabilise at extreme temperature and humidity values which means it is able to reach lower values. Our detection module was finished when the laboratory time was over which means that no results can be shown.

Environment module

1. Monitoring of the temperature and humidity module

We did many experiments to understand and calibrate this module. Our module works on adapting the heating power thanks to monitoring. To understand how temperature and humidity were evolving inside our box we run different experiments. We first calibrated the PID coefficient to obtain a smooth control, then we checked what was happening with disruptions and lastly we tried to see for how long it could be working.

All the results on the environment module are done on the prototype that you can see on figure 21. Explanations around our prototype are developed in the engineering page.

box plot 1
Figure 21: prototype for the environment module experiences

Calibration of the PID coefficient

The PID coefficients are in total 3: Kp, Ki and Kd. Basically, the time to reach the desired temperature depends on Kp but a final error and an overshoot are created, Ki lowers the final error and Kp lowers the overshoot. Hence, you have to find which triplet gives you the evolution you want. In our case, we want low overshoot and low final error because we precisely need to reach the desired temperature.

To do so, we look at the datas of sensors 0 to 4 during 30 minutes. The desired temperature is 37°C and the desired relative humidity is 100%. On the next graphes, you will see the typical shape of our curves. Temperature rises sharply and stabilises most of the time with an overshoot. Humidity rises and encounters a drop for a small time.

On the sensor number 0, you may see one or two vertical lines, this means that an error of lecture has been detected and the software puts the value at 0. We had trouble with this one sensor and later we replaced this sensor by a new one.

graph with Kp=50, Ki=0, Kd=0

This experience tells us that it is possible to stabilize our box at the desired temperature and desired humidity with only Kp=50. It needs about 15 minutes to reach the desired atmosphere and then it is stable. When you look around the plate (sensors number 0 and 1), you see a nice overshoot of about 1°K and after this overshoot there is a little slamp. It is a secondary small oscillation that can be clearly checked on raw datas.

graph with Kp=50, Ki=0.01, Kd=10

This experience shows the impact of the three coefficients together: oscillations are erased, the overshoot is lowered and temperature is stabilized sooner than the previous experience. Both curves have the same shape, but the evolution is smoothed. This triplet of coefficients are fitted for our prototype and now we will make experiences with this PID coefficients. We calibrated our PID corrector.

Evolution inside the covered multiwell plate

The next step is to understand how temperature and humidity evolve inside the covered plate. Hence, we inserted the sensor number 5 inside a plate and put the cover on. Sensors number 0, 1, 2, 3 and 4 are at the same position, so the sensors 0 and 1 are above the cover.

graph with Kp=40, Ki=0.005, Kd=10
Figure 22: results of the automation of the environment module

On the figure 22, we used the sensor number 5 to monitor what is happening inside the covered plate. Sensors number 0 to 4 measure the atmosphere in the very same way and you have the same results even though we slightly changed the PID coefficients. This slight change shows that there is a wide range of coefficient values for having a nice calibration.

Sensors number 5 measures the parameters inside the covered plate. It shows that temperature and humidity slowly catch back the parameters of the atmosphere. It needs 30 minutes for the wells to reach the temperature desired value and humidity was about 90%. Thanks to this experiment, we can validate that our incubator fulfills its first goal, which is to create the bacteria growth conditions even with the cover on.

Disruption experiments

The goal of this experience is to validate that our control works even with disruptions. It lasted 1 hour and 30 minutes (5400 seconds), the desired temperature was 37°C and the disruption was to open the cover for 10 seconds. We did this disruption at 30 minutes, 40 minutes, 70 minutes and 80 minutes. Sensor number 5 is still inside the covered plate. The figure 23 shows the result of the monitoring of this experience.
Sensor number 0 has been removed because it broke and so we adapted the software to do without its datas.

temperature evolution with perturbation
Figure 23: Experiments with disruptions

The first result is that temperature is nicely stabilised at 37°C and sometimes you see a sudden drop that matches the time when the box was opened: the heated atmosphere inside the box was leaking out and mixing with the external atmosphere at 26°C, which means our atmosphere was cooling. The consequence is the presence of variations in the temperature curves. These variations are also visible on the PWM graph (which is the heating resistance supply in percent). When temperature goes down, the PWM goes up to compensate, this is clearly shown by the sharp peaks on this graph. It takes around 5 minutes to our box to re-stabilise perfectly at the desired temperature.
You can see that the sensor number 5 inside the covered plate also receives the same variations and stabilises the same way.


An incubator should provide a closed and controlled atmosphere, work as long as you want and stabilise quickly after you open the door for a little time. We made long experiences during which we created disruptions at different moments. Results are that our prototype stabilises at the desired temperature within 20 minutes, it adapts against disruptions and it can work for 12hours long and even more.

You have to know that the perfect calibration depends on the atmosphere around the box and this variable moves by only heating or refreshing around the box. It won’t have a significant impact on the results of our prototype if it is a small variation around the overshoot.

To solve this issue, we thought of building a black box in wood around our final hardware. It has both advantages of isolating from atmosphere variations for the environment results and isolating from light for the fluorescent results.

You may have released that sometimes, humidity is not reaching 100% and is not homogeneous even though every sensor has datas higher than 90%. More time or more atomizers are needed to achieve the 100% relative humidity. For our prototype, it can reach in 30 minutes a desired relative humidity of 95%. That means that it can reach any value between the initial value and 95%. To solve issues such as failures or volume increasing, 3 atomizers have been prepared for our hardware of 36L.

2. Temperature control and bacteria growth

The chamber where the biofilms and the bacteria have to be grown is heated by resistors and monitored by microcontrollers. In order to validate this module and our construction, we carried out three experiments with the assumption that both the prototype and the industrial incubator ThermoFisher MaxQ 6000 present in the laboratory will induce the same bacteria growth.

To do so, the experiments were based on the culture of three different E. coli strains: BL21, Nissle and DH5 alpha grown in a Luria Broth medium. Each strain was placed in 8 wells of a 96 well plate as well as a control solution which contains only the LB medium. We have used these bacteria strains because they are the ones used by the team of biologists. The DH5 alpha are used for cloning, the BL21 used for the production of proteins and the Nissle is the bacteria strain which will be used at the end and will contain the entire biological system.

We prepared two identical plates, one for our prototype and the other for the industrial incubator. After starting the experiment, we followed for ten hours the optical density of each well. For bacteria growth, after 5 hours the growth isn’t as strong as at the beginning but it allowed us to validate the performances of our prototype over a long period of time. Also the bacterial growth reaches a plateau at a certain point of the development of the colonies due to the limiting access of the cells to the nutrients. In my opinion, there is nothing strange in the curves you present (except the drop of the OD of the BL21 at 9h, which might be due to an unpredicted event that disturbed the growth of the bacteria).

The optical density monitoring of the plates was carried out by FLUOstar Optima from BMG LABTECH.

bacteria population growth graph
Figure 24. Evolution of the OD of the 3 strains for 10 hours the 02/09/2020

To verify the correlation between the different evolution, we measured the distances between two curves with the spearman’s correlation coefficient. The spearman coefficient is an indicator which allows to compute a correlation coefficient for non linear correlation. The results of the calculations made with Python are as follows:

  • BL21 pearson's correlation coefficient = 0.97
  • DH5 alpha pearson's correlation coefficient = 1.0
  • ECN pearson's correlation coefficient = 0.99
  • LB pearson's correlation coefficient = 0.43

Each time, the correlation coefficients for the three strains are above 0.97 and prove that the evolution of the optical density is correlated in both incubators.

There is an exception, the correlation coefficient for the evolution of the Luria Broth medium is a bit lower, this is due to a small difference (always below 0.07) between 2 and 6 hours after the beginning of the experiment in our prototype but at the end, the OD values are equal (difference of 0.008). Plus, as there is no bacteria in the LB medium, the OD shouldn’t evolve and the small evolution can be due to some small amounts of contamination, but the evolution isn’t significant (<0.12).

Furthermore, the difference of growth between the LB and and the three other strains is significant.It can be concluded that our incubator allows us to cultivate bacteria in the same way as the industrial incubator.


    Diaz Z, Xavier KB, Miller ST. The crystal structure of the Escherichia coli autoinducer-2 processing protein LsrF. PLoS One. 2009;4(8):e6820. Published 2009 Aug 28. doi:10.1371/journal.pone.0006820