Oscillate is a novel system utilizing variable inputs to trigger variable expression. Using a combination of biological low-pass and high-pass filters, we developed a functional gene expression system which is able to respond to frequency-variable input signals.
We provide evidence that our genetic bandpass filter allows expression of GFP to occur only between a narrow range of temperature frequency, while discriminating against frequencies outside the band pass filters range.
Figure 1. System overview (a) Low pass filter with GFP (BBa_K2561003) (b) High pass filter with GFP (BBa_K2561005)(c) Low pass filter with hrpS for AND gate (BBa_K2561004)
(d) High pass filter with hrpR for AND gate (BBa_K2561006)
(e) AND gate reporter (BBa_K2561007)
We specifically designed our novel low-pass and high-pass filters to be modular, so that they could be tested and function individually without need for the full system to be intact. The effects of the individual tests run on each the low-pass and high pass filters are to be taken as oscillatory responses. Significant expression shows activation of the system and lack therein shows no response. However, expression should be cyclical and mimic the input signal.
We first confirmed the functionality of both the low pass and the high pass filter individually, by subjecting cultures with only the low pass or high pass inserted to oscillating temperatures between 37̊ C and 22̊ C for the period we were testing. Using a plate reader and taking samples of bacteria every 30 minutes, the ratio of fluorescence to bacterial cell density was plotted over time. Bacterial cell density was determined by measuring absorbance of the culture at 600 nm. Fluorescence was measured by exciting the GFP with 485 nm light and recording its emission at 510 nm.
Figure 2. The oscillatory input signal took the form of a square wave with varying period.
To account for the stability of the GFP, we report the change in the fluorescence/absorbance ratio between successive time points as the normalized fluorescence. As a result, this tends to lead to variations in the amplitude of the signal read; however, the frequency of the signal output is of importance. The plots of the high-pass and low-pass cultures resulted in an oscillating function with the low-pass and high-pass filters acting opposite each other, indicating that both parts were individually functional at frequencies within the band.
We initially tested with an oscillation period of 2 hours (138.8 μHz) to confirm the low and high pass were individually functional. This initial frequency was predicted by our deterministic model based on the approximate parameters we had determined from literature searches. Continued testing allows for the model to learn, and will further help fit parameters to values.
We continued subjecting our cultures with the individual parts in them to different frequency oscillations in temperature to find a parameter that produced the maximum fluorescent expression in both. We also tested for frequencies in which the ratio of fluorescence to bacterial cell density was inverted between the high pass and low pass cultures, so that the maxima of the high pass were aligned with the minima of the low pass and vice versa. This response is crucial for optimal bandpass functionality.
During our testing, we also tested for the upper bound (cut-off) frequencies of fluorescence for the low pass. By testing a wide range of periods, from 10 minute periods to 3 hour periods, we determined the bounds as the frequency at which the sinusoidal expression of GFP was no longer observed.
Figure 3. The high and low pass filters work opposite each other - when one peaks in expression, the others is at a local minimum.
First however, to verify that the RNA thermometer and σF were working properly, we measured fluorescence with a step-input. The results below indicate that the RNA thermometer melt temp is around 37°C, and that the sigma factor and RNA thermometer are working as intended.
Figure 4. Verification of low-pass components.
Analysis
We fit the data using Fourier analysis with the MATLAB Curve Fitting Toolbox. Using nonlinear least-squares, we determined the effectiveness of the filter by examining the angular frequency, ⍵, of the Fourier function. Given that ⍵ = 2πf, it was clear that simply fitting one term to the data was inadequate, although there was a clear correlation. Like an electronic band-pass, we would expect the frequency of the input oscillation to match the frequency of the output.
To remedy the fit, we added more terms to the Fourier series representing the output oscillations. Adding terms accounts for noise in the data by accounting for the harmonics - integer multiples of the frequency - that are overlaid. By accounting for the harmonics (second and third), we demonstrate a progressively better fit and that our system effectively passes frequencies within the band.
| Construct | Terms (in fit) | Period (T) (min) | Expected w (min^-1) | Fit w (min^-1) | R^2 | 95% Confidence Interval (w) |
|---|---|---|---|---|---|---|
| High Pass | 1 | 120 | 0.0523 | 0.1245 | 0.6491 | (0.11, 0.13) |
| High Pass | 2 | 120 | 0.0523 | 0.06166 | 0.7472 | (0.05, 0.07) |
| Low Pass | 1 | 120 | 0.0523 | 0.07642 | 0.5065 | (0.05, 0.09) |
| Low Pass | 3 | 120 | 0.0523 | 0.04002 | 0.9164 | (0, 0.12) |
| Low Pass | 1 | 40 | 0.157 | 0.1590 | 0.4248 | (0.13, 0.18) |
| Low Pass | 2 | 40 | 0.157 | 0.1765 | 0.9280 | (0.17, 0.18) |
| High Pass | 1 | 180 | 0.035 | 0.0386 | 0.3790 | (0.03, 0.05) |
| High Pass | 3 | 180 | 0.035 | 0.0435 | 0.6255 | (0.034, 0.052) |
| Low Pass | 1 | 180 | 0.035 | 0.076 | 0.6348 | (0.06, 0.08) |
| Low Pass | 4 | 180 | 0.035 | 0.0187 | 0.9050 | (0.01, 0.03) |
| Low Pass | 1 | 10 | 0.628 | 0.4439 | 0.4241 | (0.37, 0.51) |
| Low Pass | 3 | 10 | 0.628 | 0.2447 | 0.4631† | (0.12, 0.36) |
Figure 4a-4d. Low and high pass at T = 120 min. (a) Low pass with one Fourier term (b) Low pass with three Fourier terms (harmonics) (c) High pass with one Fourier term (d) High pass with two Fourier terms
Figure 5a-5b. Low pass at T = 40 min. (a) Low pass fit with one Fourier term (b) Low pass fit with two Fourier terms
Figure 6a-6d. Low and High pass at T = 180 min. (a) Low pass with one Fourier term (b) Low pass with three Fourier terms (c) High pass with one Fourier term (d) High pass with four Fourier terms
While adding terms to the Fourier series inherently improves the fit, it is important to note that the R2 is not the only important result of the analysis. From our conversation with Dr. Edwin Kan, an electrical engineer studying important biological signals, we realized that in addition to a strong coefficient of determination, we needed to ensure the signal frequency matched the input closely. If a curve we fit to the oscillatory output has a strong R2, but is not close to the correct frequency, then the signal was likely attenuated and the correlation is a result of over-fitting.
Additionally, we can attribute some of the difference in the input and output frequency to lag time in the system. If constant, this would only create a phase-shift in the output frequency; however, even small variations will influence the output oscillatory frequency.
Based on the results of the analysis, we found that the low-pass filter cutoff frequency was fH = 1633 µHz. This frequency corresponds to a period of 10 minutes. Signals at higher frequencies than the cutoff will be attenuated by our system.
We then began testing on our combined system. Beginning with the optimal frequency determined from testing of the individual parts, the ratio of fluorescence to bacterial cell density was plotted over time for the combined system. A near binary on effect was achieved, with fluorescence staying consistently high over time. Frequencies that were known to be outside the active range of either the high or low pass were tested and resulted in an “off” effect, the fluorescence was near zero and was constant over time. These results confirmed that our system was activated by frequencies within a small and specific range, while being inactive when exposed to frequencies outside this range.