Initial oxyR-oxyS-gfp system sensitivity assessment

In the oxyR-oxyS-gfp plasmid, the oxyRS regulon is activated by the oxidized form of the oxyR transcription factor, inducing GFP reporter expression and oxyR downregulation. After transforming the oxyR-oxyS-gfp in E. coli DH5a, fluorometric assays were performed. The GFP fluorescence signal was aquired over several hours, at t=0 H₂O₂ was added in different concentrations (TECAN Infinite® M Nano+ 96-well plate reader). For all further graphs, RFU (Relative Fluorescent Unit) is calculated as fluorescence normalized by OD₆₀₀, and RRFU is the relative RFU compared to the 0 mM H₂O₂ RFU.
An increase in GFP expression was measured after addition of H₂O₂, for all concentrations (Figure 1). The intensity of the GFP signal correlates positively with the concentration of H₂O₂, a higher concentration inducing a stronger signal. An initial exponential increase is observed, where the oxidative stress is strong, followed by a linear weaker increase after 1 hour, translating the stabilizaton of the cell system. As expected, the bacteria inoculated with no H₂O₂ only showed a baseline fluorescence and no increase over time.
This shows our reporter system is relevant in monitoring H₂O₂-induced oxidative stress in bacteria over time. Here, we see that even low H₂O₂ concentrations induce high GFP signals, expected from ultra-sensitive native oxidative-stress responsive systems, which confirms the need to desensitize our oxyR-based sensing system.

To assess robustness of the system, we tested a second stimulation after 2h to see if bacteria was still susceptible to oxidative stress or if durable tolerance was to be forseeable.
Visible in Figure 3, late H₂O₂ addition (+2h) leads to a second increase in fluorescence, albeit not as steep as the initial peak. The curve flattens out as intracellular H₂O₂ is depleted and the oxyS promoter does not express much GFP anymore. GFP is degraded slowly, but is probably constantly produced due to oxidative stress from nutrient deprivation in PBS.

Placing a iGEM-tested constitutive promoter in front of OxyR, thereby stopping the negative feedback-loop

To obtain a stable signal of infection, we want a stable expression of our output protein. To bypass oxyR’s negative impact on its regulation, we redesigned the promoter region to include a constitutive promoter for oxyR (See Design & Engineering page). After successful transformation of the constructed oxyR-J23101-PL-oxyS-gfp "new oxyR" plasmid into E. coli Nissle (confirmed by sequencing), we compared it to the oxyR-oxyS-gfp in E. coli Nissle. We used Nissle as this strain was the most relevant recipient for our human health application.

As seen in Figure 4, the new oxyR plasmid did not show any GFP expression, being indistinguishable from Nissle without a plasmid. This may be due to loss of plasmid between sequencing and TECAN measurement (storage at -20°C) or an issue in the plasmid design, leading to no GFP expression. The baseline GFP fluorescence at 0 mM H₂O₂ can be attributed to oxyS promoter leakiness. Finally, we can notice the reduced fluorescence for the 10 mM H₂O₂ concentration compared to 5 mM, which is most likely due to cell death under high oxidative stress. In practice, we would hope for the population to be fit enough to withstand such concentrations, as many bacterial microbiota already do by tolerance and adaptation.

Random mutagenesis of the oxyR-oxyS-gfp plasmid

To raise the ROS concentration threshold needed for GFP expression, as we need a high-pass ROS filter, we aimed to mutate the promoter region by error-prone PCR (epPCR, see Design & Engineering page). Many colonies were picked from each epPCR and characterized. Figure 5 shows the reduced GFP expression of colony 2 from the 0.6 mM MnCl2 epPCR. Results show overall decreased response but no change in concentration-specific behaviours. Sequencing revealed that error-prone-PCR did not yield any mutations in the picked colonies. We hypothesize that stress from Gibson Assembly and transformation led to transcriptional changes impacting ROS-sensitivity and our sensor system.

Further colonies were picked from the epPCR plate and measured, again without actual sequence mutations, and also demonstrating different ROS-response profiles as shown in Figure 6. Even though the Taq polymerase used has no proofreading ability and its error-rate is increased by high manganese concentration, its error-rate may still be too small for the short targeted promoter sequence, leading to no mutations.
The variation in phenotype being small and not high-pass -like, we chose to put random mutagenesis to the side, as we did not manage to generate mutants.

Directed mutagenesis of the oxyR-oxyS-gfp plasmid

In order to tune our system to be a high-pass filter of ROS, i.e. getting activated from 1mM H202 (ideally 5mM for plant infections) while conserving high output expression, we turned to computational modeling to determine relevant targeted point-mutations which would make the system better fit these criterions. The model was based on local decrease of transcription factor (OxyR) binding affinity to the promoter. We ran our model on both designed ROS-sensing systems, e.g. the oxyR-dmoxyS_OG-gfp and the oxyR-dmoxyS_BS-gfp (with a duplicated binding region(=BR)) An array of mutation combinations for both promoter versions (BR duplicated or not) was designed, cloned and transformed into DH5-alpha. OG and BS mutations respectively (8 mutated versions each) were predicted to have slight to strong impact on OxyR binding affinity to its promoter (See correspondingDesign section and Model page for details.)
In Figure 7, we represented the most relevant constructs, which show significantly different ROS-responsiveness characteristics. For instance, OG-4 shows improved GFP expression at 10 mM H202 compared to the unmutated control, but the best improved construct would, in fact, be OG-3 since the response is still high at high concentrations, but the background is more than twice as low for low concentrations, which is what we were looking for in our high-pass filter designs. As expected, versions with too many mutations showed impaired ROS-responsiveness, as seen for OG-1 (4 soft mutations). These results also show the importance of experimental support and verification of computational models, as predicted soft mutations did not induce the same impact on ROS-sensing, as seen comparing OG-2, OG-3, OG-4 and OG-5 (all with a single soft mutation). Here, we also show the success of the design and engineering of our BR-duplicated version , oxyR-dmoxyS_BS-gfp, where BS-4 more specifically shows promising high-pass filter characteristics.

In order to better analyze our data and for it to be relevant to our end application perspectives, we represented the ratios of RFU for 5 mM compared to 1 mM H202. Indeed, for plant infection detection, we would need our system to only be activated from 5 mM (ROS levels observed during fireblight infection) and inactive around 1 mM (baseline variations expected due to other abiotic stresses). Here, we confirm OG-3 being the best improved version with a 40% increase in fluorescence (Figure 8). Further experiments should be done for OG-2, since we have observed sky-rocketting fluorescence and, for this experiment, had to dilute it 1:10 so the signal could be properly detected without saturation.

Assessing host-induced variability of ROS-sensing

To adapt our ROS sensor to our different applications, our system had to show high-pass ROS-filter -relevant characteristics in our end application host strains, e.g. E.coli Nissle for IBD treatment and yeast or plant bacterial strains for fireblight detection. In the results shown Figure 9, we see how the choice of host organism is crucial to take into account as it highly impacts the ROS-responsiveness profiles. Most interestingly, E.coli K-12 and Nissle show decreased GFP-induction at lower concentrations while keeping high expression at high concentrations (from 2 mM for K-12 and 5 mM for Nissle) compared to DH5a, which is in favour of our high-pass filter objective. However, in humans, 5 mM H₂O₂ is not typically reached, and this raises the question of the pertinence of Nissle chassis for ROS-sensing in humans, as it would need further adjustements to rather respond strongly from 1 mM. Finally, we also tested high-ROS tolerant strain like the Pseudomonas alloputida mt-2 KT244 (1320), as we thought we might observe tolerance at lower concentrations and response only at high ROS concentrations (high-pass), but the 1320 strain seemed to be broadly tolerant to oxidative stress and, thus, not react at all. The 1243 strain does show high-pass behaviour, but the signal obtained at high-ROS could be too low for drone detection.

Testing the catalase output in oxyR-oxyS-gfp system (IBD model)

We are aiming to develop a theragnostic system, which associates high ROS detection to local production of anti-oxidant enzymes to counteract the ROS imbalance. Indeed, in our chosen human disease case-study, IBD, ROS are at the center of a perpetual positive feedback loop, where inflammation triggers more ROS which triggers more inflammation.
Therefore, we designed a construct with the anti-oxidant enzyme catalase as a supplementary therapeutic output to the GFP reporter, called oxyR-oxyS-KatG-gfp (See the Design & Engineering page for more details). We chose not to express the catalase constitutively but link it to ROS sensing in order to preserve bacteria fitness, mimicking native ROS pathways which produce anti-oxidant enzymes only when needed.

Catalase converts H₂O₂ into dioxygen and water. To assess its effectiveness in degrading ROS in our model assay, we tested to different quantifications, one direct through intracellular ROS titration, and the other indirect through RFU monitoring of our reporter GFP. Sadly, we did not manage to make the intracellular ROS-titration kit work in our timeframe, so we will only show our RFU data. In Figure 10, we observe significantly lower fluorescence levels and increase for the KatG plasmid version compared to the control oxyR-oxyS-gfp. Overall lower expression levels were expected since 2 genes are now under the same promoter, but the slower increase also suggests less activation of the promoter, so less oxidative stress and thus elimination of H₂O₂ molecules. Our interpretation is that catalase expression is successfully induced and participates in regulating the H₂O₂ levels, thereby keeping oxidative stress low, in turn leading to less reporter GFP expression. These results show the promise of our theragnostic platform for IBD treatment, although more direct titrations are necessary to be able to validate the hypothesis above.

Our sensor tested in yeast

Cloning the Trx2-gfp constructs

For the plant application, we developped in parallel of our bacterial platform, our yeast platform, since yeast have shown to be effective as biocontrol agents. Instead of the bacterial ROS-sensing TF OxyR, we used the native yeast TF YAP1 and the plant TF TGA2 (expressed in yeast), designed as yap1-trx2-gfp and Tga2-CaMV35S-gfp constructs (See Design & Engineering page). Here, we show only YAP1 data, as TGA2 did not show any response and we think co-activators present in the plant should have been co-expressed in yeast for the system to work.

As a reminder, trx2 is the promoter region bound by the oxidative-stress regulated YAP1 TF naturally present in S. cerevisiae. We did clone the trx2-gfp plasmid but did not manage to insert the yap1 gene. Since YAP1 is natively expressed, we could still test the trx2-gfp construct. The trx2-gfp showed nearly a 3-fold increase in RFU from 1 mM H₂O₂ and background fluorescence below, which would be perfect for the human application but, for the plant application, we would need activation only from 5 mM H₂O₂. These results are very encouraging, trx2-gfp exhibiting a switch-like behaviour. Further, we would also do directed mutagenesis on the trx2 promoter in order to shift the activation threshold from 1 mM to 5 mM H₂O₂ and obtain our ideal high-pass filter.

Solving spectra overlap issues

GFP exhibits an emission wavelength which overlaps too much with the autofluorescence of yeast and plants, so a reporter gene was selected to replace GFP, ymScarlet, which would be compatible with observed autofluorescences.
After our first cloning (digestion and ligation), we observed no ymScarlet emission (λ=594) for mScarlet-X0 "V1", as shown Figure 12. This probably results from design of the ligation, where the 5' and 3' distances after the promoter and before the terminator were changed slightly, thus leading in expression hindrance.
We tried again through Gibson Assembly (amplifying the backbone). This time, mScarlet-X0 "V2" showed very high ymScarlet signal, demonstrating the success of our design and engineering, and our DESIGN-BUILD-TEST cycle approach.
Due to time restrictions, this process was not yet repeated for all other constructs.

Conclusion & outlook

We have successfully engineered mutant variants of the OxyR-based sensor system which show improved high-pass sensor characteristics (see OG-3 Figure 8). Further tuning and characterizations would allow for more knowledgeable selection of a good candidate to pursue the process of developping the theragnostic system with. Indeed, further titrations are needed to properly validate the effectiveness and mode of action of the catalase output (Figure 10), and should be put under the control of the selected improved oxyR promoter. Further down the line, for biosecurity reasons and stability purposes, the sensing system should be genomically integrated in the bacterial/yeast DNA. Genomic integration will surely impact ROS-responsiveness and the system will have to be carefully re-characterized. For IBD treatment, a kill-switch should also be integrated.
We noticed non-replicability of assays using the same colonies from the same cryostocks, where cells seemed to lose ROS-responsiveness and we think lost their plasmid during storage, since the stocks were mistakenly stored at -20°C rather than -80°C. This left us unable to repeat some characterizations under standardized conditions. We would retransform some of the plasmids to generate more reliable and standardized data, especially for the yeast constructs.
In yeast, we obtained very promising results for the trx2-gfp construct, exhibiting a switch-like behaviour, with an distinct activation from 1 mM H202 (Figure 11). In order to shift the activation from 1 mM to 5 mM H202 (relevant range in plants), we would do directed mutagenesis on the trx2 promoter, in the same way as for the oxyR-based sensor. Ideally, in yeast, for better measurments and end-application compatibility, we would also switch all GFPs to ymScarlet, avoiding overlaps with natural yeast and plant autofluorescence ranges.
Future plans include transforming Pf-5 bacterial biocontrol strain for the plant application, as an alternative to yeast.
All constructs would be tested in a plant model in vitro. We had planned to co-inoculate a cut flower blossom with the pathogenic and engineered sensor bacteria/yeast at Agroscope, but did not have suitable candidates to test yet.