Results

   Summary

We characterized more than two dozen parts over the course of this project. Here, we focus our discussion on the biosensor constructs that best illustrate our successful proof-of-concept experiments. These parts are [lasP+RBS30+GFP, CP+RBS30+lasR], [rhlP+RBS30+GFP, CP+RBS30+rhlR], and [GR(S)+RBS34+GFP, CP+RBS34+RhlR].

In order to evaluate the suitability of our biosensor constructs for detecting P. aeruginosa, we conducted a series of dose-response studies for characterization purposes. We also analyzed this data to determine the transfer function and dynamic range of each biosensor system.

Figure 1: Dose response of the PAI-1 biosensor system (LasP+RBS30+GFP and CP+RBS30+lasR). Immediately before the assay, cells were diluted to ensure that they were growing in the exponential phase for the experiment. Autoinducer PAI-1 was added at the concentrations indicated, and GFP fluorescence was quantified using an incubated, shaking plate reader. In this plot, fluorescence was normalized to the culture OD to control for cell growth. Samples were run in quadruplicates with standard deviation indicated (error bars = SD; n=4).

Our first observation was that the construct [lasP+RBS30+GFP, CP+RBS30+lasR] appears to be well-suited as a binary sensor and can be used to simply determine whether or not P. aeruginosa is present in a sample. The construct follows the same general trend of fluorescence at varying autoinducer concentrations, except for the negative control in which no autoinducer was present. There is a significant amount of overlap in the error bars in Figure 1, so T-tests were conducted in order to evaluate the statistical significance of this observed trend, as detailed below in Table 1.

Table 1: Statistical analysis of the PAI-1 biosensor response (lasP+RBS30+GFP, CP+RBS30+lasR). The null hypothesis is that there is no statistical difference between the means of the compared samples. Green cells indicate rejection of the null hypothesis (p<0.05), while blue cells indicate failure to reject (p>0.05). (A) Data from each of the initial segments of the curves (region before any fluorescence is observed ~30min) was compared between varying autoinducer concentrations. (B) Data from the initial segment of each curve (before any fluorescence is observed) was compared with the final steady state fluorescence of that curve from the same sample (last 10 data points). (C) Data from each of the final steady state segments of the curves was compared between varying autoinducer concentrations (last 10 data points).

In the first 30 minutes after autoinducer dosages were given, only the sample that was given 100μM of PAI-1 exhibited a response that was significantly different from the other samples (Table 1A). Table 1b tests the activation of each construct. This is accomplished by comparing the data from the initial segment of the curves (before any fluorescence is observed) with the final steady state fluorescence (last 10 data points). Our data also indicated that fluorescence per OD changed significantly over time in each sample (Table 1B).

Table 1C compares the data from each of the final steady state segments of the curves with the other final steady state segments (last 10 data points). The negative control exhibited significantly different fluorescence readings per OD. This is because the cells showed relatively steady total fluorescence coupled with rapid of cell growth, resulting in a decrease in fluorescence per OD, as shown in Figure 2 below. All the samples showed similar fluorescence readings except for the 0.1μM, to some extent the 0.5μM, and of course 0μM PAI-1 negative control.

Figure 2: Overall response of the PAI-1 biosensor system (lasP+RBS30+GFP, CP+RBS30+lasR). Here, the data from Figure 1 are presented without normalization, showing total fluorescence (left) and total OD of the cultures.

In order to further characterize this biosensor, we next plotted the steady state fluorescence (per OD) vs. autoinducer concentration to determine the input-output transfer function (Figure 3). As indicated by the analysis and discussion above, in this “binary” biosensor, fluorescence per OD stayed relatively constant (within about 10% of the mean fluorescence per OD) for all samples treated with PAI-1 autoinducer, and all samples exhibited significantly different behavior in contrast to the negative control sample.

Figure 3: Input-output transfer function for PAI-1 biosensor (lasP+RBS30+GFP, CP+RBS30+lasR). Steady-state responses were calculated from the data in Figure 1 and plotted against the input concentration of PAI-1 autoinducer.

In contrast to the binary detection system, the construct [rhlP+RBS30+GFP, CP+RBS34+rhlR] is well suited for determining the amount of P. Aeruginosa present in a sample because it appears to be able to detect and discriminate between varying autoinducer concentrations. Figure 4 below details the fluorescence per OD observed upon the induction of the system at multiple autoinducer concentrations. Unlike the binary detection system, the fluorescence per OD of each curves is distinctive.

Figure 4: Dose response of the PAI-2 Biosensor System 1 (rhlP+RBS30+GFP, CP+RBS34+rhlR). This experiment was conducted as in Figure 1, and strains were exposed to 0 - 100 micromolar PAI-2. Fluorescence and OD were measured every 7.5 min (error bars = SD; n=4).

For the most part, error bars only overlapped when the autoinducer concentration exceeded 15μM, which is not a physiologically relevant value. Notably, autoinducer concentrations between 0μM and 5μM are clear and distinguishable. T-tests were conducted in order to statistically confirm the variance of each concentration curve, as detailed below in Table 2.

Table 2: Statistical analysis of the PAI-2 Biosensor 1 response (rhlP+RBS30+GFP, CP+RBS34+rhlR). The null hypothesis is that there is no statistical difference between the means of the compared samples. Green cells indicate rejection of the null hypothesis (p<0.05), while blue cells indicate failure to reject (p>0.05). (A) Data from each of the initial segments of the curves (region before any fluorescence is observed ~30min) was compared between varying autoinducer concentrations. (B) Data from the initial segment of each curve (before any fluorescence is observed) was compared with the final steady state fluorescence of that curve from the same sample (last 10 data points). (C) Data from each of the final steady state segments of the curves was compared between varying autoinducer concentrations (last 10 data points).

Table 2A, as before, compares the early time points, before GFP is induced (t=30 min). As we observed with the PAI-1 biosensor, only the 100μM sample significantly induced fluorescence in the PAI-2 Biosensor 1 within the first 30 minutes.

Table 2B, as before, shows promoter activation, by comparing the data from the initial segment of the curves (before any fluorescence is observed) with the final steady state fluorescence (last 10 data points). The system exhibited behavior that is similar to the behavior of the binary sensor, as each construct changes a statistically significant amount when in the presence of autoinducers. Surprisingly, the 0μM autoinducer sample also has a statistically significant change. However, the induced constructs produced fluorescence readings that were orders of magnitude higher than the negative control, so any cell bias introduced by an OD irregularity is insignificant and can be ignored.

Table 2C compares the data from each of the final steady state segments of the curves with the other final steady state segments (last 10 data points). The data shows that the final steady-state fluorescence of nearly every sample is distinct in comparison to the final-steady state fluorescence of samples that were administered a different autoinducer concentration. The two exceptions are the 7.5μM-10μM, and the 15μM-20μM autoinducer concentration samples. It is important to note that 10μM-15μM and 20μM-50μM concentration samples have statistically significant responses. In fact, the high degree of discrimination between relative autoinducer concentrations strongly qualifies this construct as a useful concentration-based detection method. The steady state fluorescence per OD is presented in Figure 5. The logarithmic regression fits the data quite well, and proves that the construct can be used in a biosensor application.

Figure 5: Input-output transfer function for PAI-2 Biosensor 1 (rhlP+RBS30+GFP, CP+RBS34+rhlR). Steady-state responses were calculated from the data in Figure 4 and plotted against the input concentration of PAI-2 autoinducer.

In addition to designing plasmid systems using existing biobricks from the registry, we also engineered a plasmid that contains an inducible promoter that was isolated from the genome of P. aeruginosa. Out of all the parts that we derived from the P. aeruginosa genome, the construct [GR(S)+RBS34+GFP, CP+RBS34+RhlR] was most suitable in autoinducer detection. Figure 6 shows our GR(S) promoter is sensitive to a range of autoinducer concentrations and requires an average of two hours for reach full induction.

Figure 6: Dose response of the PAI-2 Biosensor System 2 (GR(S)+RBS34+GFP, CP+RBS34+RhlR). This experiment was conducted in a similar fashion as the experiment shown in Figure 1, and strains were exposed to 0 - 100 micromolar PAI-2. Fluorescence and OD were measured every 7.5 min (error bars = SD; n=4).

Just as in the previous concentration-dependent detection system, the genomic promoter system also discriminates very well between the varying autoinducer concentrations. To confirm our results, we conducted T-tests to validate the observed significance between the curves. The T-tests conducted follow the same format as the prior ones described above and is detailed below in Table 3.

Table 3: Statistical analysis of the PAI-2 Biosensor 2 response (GR(S)+RBS34+GFP, CP+RBS34+RhlR). The null hypothesis is that there is no statistical difference between the means of the compared samples. Green cells indicate rejection of the null hypothesis (p<0.05), while blue cells indicate failure to reject (p>0.05). (A) Data from each of the initial segments of the curves (region before any fluorescence is observed ~30min) was compared between varying autoinducer concentrations. (B) Data from the initial segment of each curve (before any fluorescence is observed) was compared with the final steady state fluorescence of that curve from the same sample (last 10 data points). (C) Data from each of the final steady state segments of the curves was compared between varying autoinducer concentrations (last 10 data points).

Table 3A shows that the genomic promoter construct had better uniformity in early time points than our concentration detection system (discussed above). The initial data of almost every sample coincided with that of the negative control (0μM). Once again, the 100μM sample exhibited induced fluorescence quickly. Thus, virtually all the samples exhibit fluorescence similar to that of the negative control at this short time point.

In Table 3B, the data shows that similar to previous systems, each construct exhibited a statistically significant amount of fluorescence per OD in the presence of autoinducers. The only outlying T-test in this characterization is the 0.1μM-20μM comparison, which may indicate that the uninduced state of the 20μM autoinducer sample was close to the induced state of the 0.1μM autoinducer sample’s fluorescence.

Table 3C's data validates the claim that the fluorescence per OD of every single steady state curve is statistically different across varying autoinducer concentrations. This result is further highlighted by the steady state fluorescence per OD vs. autoinducer concentration in Figure 8. The low error and an extremely well fit logarithmic regression exhibited proves that the construct can not only works exactly the way it should, but is also the most suitable construct in a biosensor application.

Figure 7: Input-output transfer function for PAI-2 Biosensor 2 (GR(S)+RBS34+GFP, CP+RBS34+RhlR). Steady-state responses were calculated from the data in Figure 6 and plotted against the input concentration of PAI-2 autoinducer.

In the modelling section, the theoretical math behind the system was discussed. In order to fully understand our system, the mathematical model needed to support the experimental data. Consequently, a sensitivity analysis was conducted (also in the modelling section) to gauge which factors influenced the system the most. A number of parameters were found to exert a significant amount of influence on the system. Each of these parameters was carefully adjusted until the theoretical graphs closely resembled those received from experimental testing, as demonstrated below in Figure 8.

Figure 8: Curve fit to the experimental data using the math model

Upon administration of autoinducer molecules to a sample, the intracellular concentration of the autoinducer increases dramatically as it is passively tranported through the cell membrane. Similarly, the initial concentration of R-proteins is quite high relative to other biochemical species in the cell because it is constitutively produced. However, the increase in intracellular autoinducer concentration facilitates an instantaneous drop in the (free) R-protein concentration. The R-protein/dimer concentration level increases almost simultaneously, inducing the R-protein/autoinducer-induced promoter. The rise in the dimer complex facilitates a sudden increase in expression of the reporter construct, producing GFP. The concentration of GFP increases steadily due to the increasing concentration of the dimer complex.

Autoinducer concentrations continue to drop until the reversible binding of the R-protein, autoinducer, and dimer complex reaches steady state. This approximately happens 150 minutes after induction. At this point, most of the autoinducer is bound to the dimer complex, and any remaining amount is degrading. In a way, the dimer complex is protecting the autoinducer from degradation, which maintains steady R-protein and dimer complex levels. Meanwhile, GFP is generated due to the initial excessive amount of reporter expression. However, when the R-protein and dimer complex reach steady state, GFP expression reaches steady state as well. The model fits the fluorescence data in the above characterization graphs. In the GFP curve, the concentration actually decreases slightly over an extended period of time, which is exactly what was observed experimentally.

Additionally, the critical parameters were varied until the model coincided with the experimental data, as shown in Figure 1, 4, and 6. Consequently, the model's application to the LasR system can be seen below in Figure 9. As one can observe, the model qualitatively represents the data. Almost all of the autoinducer dosages produce the same concentration of GFP.

Figure 9: Curve fit to the LasR experimental data (Figure 1) using the math model

Moreover, Figure 10 illustrates the model applied to fit the RhlR experimental data (Figure 2). Note the difference between the scale of the y-axis scale of Figure 9 and Figure 10. This model shows that the system discriminates between varying autoinducer dosages, just as the response exhibited by the actual RhlR construct in experimental findings. The y-axis range here is almost one order of magnitude higher than the LasR system, which is also supported by the data.

Figure 10: Curve fit to the RhlR experimental data (Figure 1) using math model

The difference in parameter values facilitate the binary and dosage-dependent responses of the Las and Rhl systems. Parameters that primarily contribute to this observed difference are depicted below in Figure 11.The degradation rates of RhlR were found to be much lower than that of the LasR. The LasR/PAI-1 dimer was also found to possess a higher degradation rate than its RhlR/PAI-2 counterpart. Furthermore, it is likely that the RhlR dimer exhibited positive cooperativity when binding to its cognate induced promoter to an extent. These factors are the main culprits behind the resultant dramatic changes observed in the biosensor response.

Figure 11: Critical parameters that influence the difference between the Las system (binary response) and the Rhl system (dosage dependent response). The red circles indicate the critical parameters which vary between the Las and the Rhl system. The two figures on the right demonstrate the Las (Figure 1) and Rhl (Figure 4) systems.

On the whole, we demonstrated that our biosensor constructs are functional and are excellent candidates for a cell-based biosensor system. Both the Las and Rhl reporter constructs exhibited fluorescence that was statistically significant relative to the controls at most or all of the autoinducer concentrations we tested. Compared to the Las (PAI-1) construct, the Rhl (PAI-2) system produced more fluorescence overall and output fluorescence readings that were more dependent on the concentration of autoinducers present (potentially indicating a broader dynamic range for these biosensors). The Las constructs produced very similar fluorescence values at almost every autoinducer concentration. Our results provide detailed characterizations of not only the original registry Las and Rhl promoters on which our constructs are based, but also of the newly-assembled biosensor constructs we built. Our results provide a solid basis for a synthetic cell-based “device” that can detect the presence of Pseudomonas aeruginosa.

Our RFP constructs were not as successful. The results were inconsistent and the signals did not seem to be correlated to the presence of the autoinducers. Although we were not able to determine the cause of this problem, we have put the parts in the registry for future investigation.

In addition to working with the Las and Rhl dependent promoters already on the registry, we also characterized some new promoters isolated directly from the Pseudomonas genome. This resulted in another potential set of promoters for use in a Pseudomonas detection device. In particular, the Rhl-dependent promoter we extracted from the genome exhibited excellent sensitivity to PAI2.

Moreover, we identified the critical parameters in our system which contribute to the observed fluctuations in the biosensor response. By using our math model, we can take the constructs we built and can tune the sensitivity of the biosensor to a specific range of autoinducer dosages through varying the critical parameters. This gives the system a high level of versatility to adjust the biosensor towards specific types of detection applications.

Finally, we built a generalized model which represents autoinducer detection. This model can be applied to any number of other genetic constructs designed to detect pathogens. In our system, an autoinducer is detected, and a resultant induced promoter produces a fluorescent protein to facilitate detection. To apply the system design to another pathogen, one only needs to identify the associated autoinducers, the cognate R-proteins, and an inducible promoter. Once the construct is built, it can be analyzed using the mathematical model and geared towards whichever specific system that a research is attempting to characterize.

For complete characterization information on all of our parts, please refer to our Biobricks page. Clicking on a part will take you to the Registry of Standard Biological Parts, where the complete analysis for that part is posted. Additionally, our testing protocol can be found here.