Our approach to modelling our growth switch and growth knob control systems involved the following steps:
- a) Preliminary Simulations
- b) Curve-fitting
- c) Sensitivity Analysis and Predictions
- i) the HicA-HicB toxin-antitoxin system (growth switch) - which has the ability to switch the bacterial cells on and off.
- ii) the SgrS glucose uptake system (growth knob) - which fine-tunes bacterial productivity.
Fig. 1: Proposed mechanism of the HicA-HicB toxin-antitoxin system.
- 1) All inducible and repressible systems were assumed to follow Hill equation characteristics.
- 2) Initial concentration of all species inside the cell was assumed to be 0.
- 3) The initial value of \(\mathit{OD}_{600}\) was assumed to be 0.2.
- 4) The production of HicA toxin was assumed to be unaffected by its effect on the global translation (Fig. 1).
- 5) The degradation of the HicA-HicB complex was considered negligible compared to its dissociation.
Symbol | Description |
\([\mathit{HicAmRNA}]\) | Concentration of HicAmRNA present in the cell |
\(\mathit{syn_{HicAmRNA}}\) | Rate of transcription of the HicA gene, associated with the pLac promoter |
\([\mathit{IPTG}]\) | Concentration of the inducer (IPTG) used to induce the transcription of HicAmRNA |
\(k_{\mathit{IPTG}}\) | Concentration of IPTG required to achieve half of the maximum synthesis rate of HicAmRNA |
\(\mathit{deg_{HicAmRNA}}\) | Rate of degradation associated with HicAmRNA |
\([\mathit{HicBmRNA}]\) | Concentration of HicBmRNA present in the cell |
\(\mathit{syn_{HicBmRNA}}\) | Rate of transcription of the HicB gene, associated with the pBAD promoter |
\([\mathit{Ara}]\) | Concentration of the inducer (Ara) used to induce the transcription HicBmRNA |
\(k_{\mathit{Ara}}\) | Concentration of Arabinose required to achieve half of the maximum synthesis rate of HicBmRNA |
\(\mathit{deg_{HicBmRNA}}\) | Rate of degradation associated with the HicBmRNA |
\([\mathit{HicA}]\) | Concentration of HicA toxin protein present in the cell |
\(\mathit{syn_{HicA}}\) | Rate of translation of HicAmRNA, associated with rbs34 |
\(\mathit{deg_{HicA}}\) | Rate of degradation associated with the HicA protein |
\([\mathit{HicB}]\) | Concentration of HicB antitoxin protein present in the cell |
\(\mathit{syn_{HicB}}\) | Rate of translation of HicBmRNA, associated with rbs34 |
\(\mathit{deg_{HicB}}\) | Rate of degradation associated with the HicAB protein |
\([\mathit{complex}]\) | Concentration of HicA-HicB complex present in the cell |
\(k_a\) | Association constant for toxin-antitoxin complex formation |
\(k_d\) | Dissociation constant for toxin-antitoxin complex breakdown |
\(\mu_{\mathrm{max}}\) | Maximum growth rate of the E.coli strain in the absence of induction |
\(\mathit{OD}_{600_{\mathrm{max}}}\) | Final OD600 value achieved by the E.coli strain in the absence of induction |
\(k_{HicA}\) | Concentration of HicA toxin required to achieve half of the maximum repression of growth rate |
Fig. 2: Simulations of bacterial growth with varying inducer (IPTG) concentrations used to induce HicA toxin.
Fig. 3: Simulations of bacterial growth with varying inducer (Ara) concentrations used to induce HicB antitoxin.
Fig. 4: Illustration of the mechanism of action of SgrS.
- 1) The kinetic steps listed above are described with rate equations, while the inducible transcription of SgrS in our system is modelled by a Hill equation.
- 2) The limiting substrate for growth in our system has been assumed to be glucose.
- 3) The initial \(\mathit{OD}_{600}\) is assumed to be 0.1, the initial glucose concentration to be 0.2%, the initial SgrS concentration to be 0.1636 \(\mu\mathrm{M}\), the initial ptsG mRNA concentration to be 4.632 \(\mathrm{nM}\), and the initial SgrS-ptsG complex concentration to be 0.2525 \(\mathrm{nM}\).
Symbol | Description |
\([\mathit{SgrS}]\) | Concentration of SgrS RNA present in the cell |
\([\mathit{ptsG}]\) | Concentration of ptsG mRNA present in the cell |
\([\mathit{complex}]\) | Concentration of SgrS-ptsG complex present in the cell |
\(\mathit{syn_{SgrS_{\mathrm{endo}}}}\) | Endogenous rate of transcription of the SgrS gene |
\(\mathit{syn_{SgrS_{\mathrm{ind}}}}\) | Induced rate of transcription of the SgrS gene, associated with the ptet promoter |
\(\mathit{syn_{SgrS_{\mathrm{max}}}}\) | Maximum rate of transcription of the SgrS gene, associated with the ptet promoter |
\(k_{\mathit{aTc}}\) | Concentration of aTc required to achieve half of the maximum synthesis rate of Sgrs |
\(\mathit{deg_{SgrS}}\) | Endogenous degradation rate of SgrS |
\(\mathit{syn_{ptsG}}\) | Endogenous transcription rate of ptsG |
\(\mathit{deg_{ptsG}}\) | Endogenous degradation rate of ptsG |
\(k_{\mathrm{on}}\) | Association constant for SgrS-ptsG complex formation |
\(k_{\mathrm{off}}\) | Dissociation constant for SgrS-ptsG complex breakdown |
\(k_{\mathrm{deg}}\) | tRNase E-mediated degradation constant of SgrS-ptsG complex |
\([\mathit{aTc}]\) | Concentration of the inducer (aTc) used to induce the transcription Sgrs RNA |
\(\mathit{OD}_{600}\) | OD600 |
\([\mathit{glucose}]\) | Concentration of glucose in the growth media |
\(\mu_{\mathrm{max}}\) | Maximum rate of cell growth |
\(k_G\) | Concentration of glucose required to achieve half of the maximum of cell growth rate |
\(k_{G_{\mathrm{max}}}\) | Maximum of \(k_G\) |
\(k_p\) | Concentration of ptsG required to achieve half of the repressive effect of ptsG on \(k_{G_{\mathrm{max}}}\) |
\(\mathit{Yield}\) | Yield coefficient, biomass per mass of glucose utilized |
Fig. 5: Simulations of bacterial growth with varying inducer (aTc) concentrations used to induce SgrS.
Fig. 6: Fitting of the preliminiary model to the HicA-HicB characterisation data, for varying inducer (ara) concentrations used to induce HicB antitoxin while a constant inducer (IPTG) concentration was used to induce HicA toxin.
Fig. 7: Fitting of the final model to the HicA-HicB characterisation data, for varying inducer (IPTG) concentrations used to induce HicA toxin.
Fig. 8: Fitting of the final model to the HicA-HicB characterisation data, for varying inducer (ara) concentrations used to induce HicB antitoxin while a constant inducer (IPTG) concentration was used to induce HicA toxin.
- 1) All inducible and repressible systems were assumed to follow Michaelis-Menten characteristics.
- 2) Initial concentration of all species inside the cell was assumed to be 0.
- 3) The initial value of \(\mathit{OD}_{600}\) was assumed to be 0.2
- 4) The production of HicA toxin was assumed to be unaffected by its effect on the global translation (Fig. 1).
- 5) The effect of HicB antitoxin on the HicA toxin has been assumed to be limited to a repression of the toxin's inhibtiory effect on growth.
Parameter | Description | Optimized Value (3 s.f.) |
\(\mathit{syn_{HicA}}\) | Maximum synthesis rate of HicA toxin | \(7.32 \times 10^{-8} \text{ } \mu\mathrm{M}\mathrm{min}^{-1}\) |
\(\mathit{deg_{HicA}}\) | Degradation rate of HicA toxin | \(4.00 \times 10^ {-3} \text{ } \mathrm{min}^{-1}\) |
\(k_{\mathit{IPTG}}\) | Concentration of IPTG required to achieve half of the maximum synthesis rate of HicA | \(0.0949 \text{ } \mathrm{mM}\) |
\(\mathit{syn_{HicB}}\) | Maximum synthesis rate of HicB antitoxin | \(1.16 \times 10^{-7} \text{ } \mu\mathrm{M}\mathrm{min}^{-1}\) |
\(\mathit{deg_{HicB}}\) | Degradation rate of HicB antitoxin | \(0.0130 \text{ } \mathrm{min}^{-1}\) |
\(k_{\mathit{ara}}\) | Concentration of Arabinose required to achieve half of the maximum synthesis rate of HicB | \(0.0700\%\) |
\(\mathit{OD}_{600_{\mathrm{min}}}\) | Parameter accounting for the inflection point in the growth curve | \(0.485\) |
\(\mathit{OD}_{600_{\mathrm{max}}}\) | Final OD600 value achieved by the E.coli strain in the absence of induction | \(1.35\) |
\(\mu_{\mathit{max}}\) | Maximum growth rate of the E.coli strain in the absence of induction | \(0.0250 \text{ } \mathrm{min}^{-1}\) |
\(l\) | Logistic coefficient accounting for gradient of the growth curve | \(0.165\) |
\(k_{\mathit{HicA}}\) | Concentration of “effective” HicA toxin required to achieve half of the maximum repression of growth rate | \(1.35 \times 10^{-6} \text{ } \mu\mathrm{M}\) |
\(n_{\mathit{Hic}}\) | Hill-coefficient for the repressive effect of “effective” HicA toxin on growth rate | \(6.06\) |
\(k_{\mathit{HicB}}\) | Concentration of HicB antitoxin required to make the “effective” HicA toxin level half of the original HicA toxin concentration | \(1.34 \times 10^{-6} \text{ } \mu\mathrm{M}\) |
Fig. 9: Fitting of the preliminiary model to the SgrS characterisation data, for varying inducer (aTc) concentrations used to induce SgrS.
Fig. 10: Fitting of the final model to the SgrS characterisation data, for varying inducer (aTc) concentrations used to induce SgrS.
- 1) The effect of the inducer on growth is modelled by a Hill equation.
- 2) The dilution and degradation of the inducer are accounted for in a rate equation.
- 3) The initial glucose concentration in the media is assumed to be 0.2% as per our experimental design, while the initial \(\mathit{OD}_{600}\) is assumed to be 0.14.
Parameter | Description | Optimized Value (3 s.f.) |
\(\mathit{deg_{aTc}}\) | Degradation rate of \([\mathit{aTc}]\) | \(2.15 \times 10^{-4} \text{ }\mathrm{hr}^{-1}\) |
\(\mu_{\mathrm{max}}\) | Maximum growth rate of the E.coli strain in the absence of induction | \(1.05 \text{ }\mathrm{hr}^{-1}\) |
\(k_G\) | Concentration of \(\mathit{glucose}\) required to achieve half of the maximum of cell growth rate | \(0.317 \%\) |
\(k_{aTc}\) | Concentration of \(\mathit{aTc}\) required to achieve half of the maximum repression of cell growth rate | \(700 \text{ } \mathrm{nM}\) |
\(\mathit{Yield}\) | Yield coefficient, biomass per mass of glucose utilized | \(4.48\) |
\(m\) | Hill-coefficient for the repressive effect of Inducer(aTc) on cell growth rate | \(0.515\) |
\(\mu\) | Rate of cell growth | - |
\([\mathit{aTc}]\) | Concentration of the inducer (aTc) used to induce the transcription of SgrS mRNA | - |
\(\mathit{OD}_{600}\) | OD600 | - |
\([\mathit{glucose}]\) | Concentration of glucose in the growth media | - |
Fig. 15: Sensitivity analysis of modifiable parameters of the model- \(\mathit{deg_{HicA}}\), \(\mathit{syn_{HicA}}\), \(\mathit{deg_{HicB}}\) and \(\mathit{syn_{HicB}}\). Sensitivity was analysed by studying the change in the growth profile from the optimized baseline model, as the different parameters were varied.
Fig. 16: Simulation of the effect of \(\mathit{deg_{HicA}}\) on cell growth. The black solid line represents the growth trend for uninduced sample; the red lines represent growth trends for induced HicA toxin with inducer (IPTG) concentration of 1mM. The solid, dashed and dotted lines are increasing values of \(\mathit{deg_{HicA}}\) , and the solid line is the optimized parameter value. A smaller \(\mathit{deg_{HicA}}\) results in greater growth arrest.
- a) Maximum arrest due to toxin
- b) Maximum recovery due to antitoxin
- c) Minimal dose of IPTG for toxin induction
- d) Minimal dose of Arabinose for antitoxin induction
Fig. 17: Response surface depicting the value of the objective functions for different concentrations of IPTG (mM) and Arabinose (%). Combination of IPTG and Arabinose concentrations in the red region as well as the surrounding blue region were tested by the wet lab team.
Fig. 18: Results of the characterisation experiment for the combinations belonging to the red and blue regions of the surface plot above. The trends observed in the experiment validated the predictions made by the model and the most optimal inducer combination was found to be 0.5mM IPTG (for toxin) and 0.4% Ara (for antitoxin).
Fig. 19: Simulations of bacterial growth with varying induction timepoints of toxin.
Fig. 20: Final OD600 values for different induction timepoints of toxin. Inducing the HicA toxin at a later timepoint results in a less-pronounced growth arrest.
Fig. 21: Simulations of bacterial growth with varying induction timepoints of antitoxin. Inducing the HicB toxin at a later timepoint results in a less-pronounced growth resumption from the arrested mode.
Fig. 22: Simulations of bacterial growth for varying inducer (aTc) concentrations used to induce SgrS. Increased inducer concentrations result in slower rate of growth, prolonging the duration of growth before reaching the final OD600 value.
Fig. 23: Dose response of the maximum rate of cell growth with inducer (aTc) concentrations used to induce SgrS. Increased inducer concentrations result in the decrease of the maximum growth rate. The maximum growth rate drops steeply when inducer concentration is increased from 0nM to 400nM.
Fig. 24: Simulations of glucose remaining in media over time for varying inducer (aTc) concentrations used to induce SgrS. Increased inducer concentrations result in slower consumption of glucose, prolonging the duration of growth before all glucose is used up.
Fig. 25: Sensitivity analysis of the modifiable parameters of the model, \(\mathit{deg_{aTc}}\), \(\mu_{\mathrm{max}}\), \(k_G\) and \(k_{aTc}\). Sensitivity was analysed by studying the change in the growth profile from the optimized baseline model, as the different parameters were varied.
Fig. 26: Simulation of the effect of \(k_{\mathit{aTc}}\) on cell growth. The black solid line represents the growth trend for uninduced SgrS; the red lines represent growth trends for induced SgrS with inducer (aTc) concentration of 1000nM. The solid, dashed and dotted lines are increasing values of \(k_{\mathit{aTc}}\), and the dashed line is the optimized parameter value. A smaller \(k_{\mathit{aTc}}\) value resulted in greater growth repression.
Fig. 27: Simulation of the effect of \(k_G\) on cell growth. The black solid lines represents the growth trend for uninduced SgrS; the red lines represent growth trends for induced SgrS at an inducer (aTc) concentration of 1000nM. The solid, dashed and dotted lines are increasing values of \(k_G\), and the dashed line is the optimized parameter value. Growth reduction due to induced SgrS (red) compared to uninduced SgrS (black) is more pronounced with a greater \(k_G\) value.
Fig. 28: Simulation of the effect of \(\mu_{\mathrm{max}}\) on cell growth. The black solid lines represents the growth trend for uninduced SgrS; the red lines represent growth trends for induced SgrS at an inducer (aTc) concentration of 1000nM. The solid, dashed and dotted lines are increasing values of \(\mu_{\mathrm{max}}\), and the dashed line is the optimized parameter value. Growth reduction due to induced SgrS (red) compared to uninduced SgrS (black) is more pronounced with a a slower growth rate (smaller \(\mu_{\mathrm{max}}\)).
Growth Switch Model | Growth Knob Model |
---|---|
|
|
Fig. 11: One-dimensional trace plots for parameters of the HicA-HicB growth switch model.
Fig. 12: Two-dimensional trace plots for parameters of the HicA-HicB growth control model.
Overall, the HicA-HicB growth switch model developed was of paramount importance to the success of our project as it provided useful insights pertaining to optimisation and predictability of the HicA-HicB growth switch . Upon deeper analysis, we found that the predictions woudl be more reliable if the identifiability aspect of the model could be improved. A way to improve the identifiability of our model is to have the wet lab team carry out experiments that allowed measurement of different variables, which would help introduce more constraints on the parameters. Another solution will be to develop an even simpler model that could still be representative of the biological system at but contain fewer unknown parameters.Fig. 13: One-dimensional trace plots for parameters of the SgrS growth control model.
Fig. 14: Two-dimensional trace plots for parameters of the SgrS growth control model.
Based on the convergence of parameter estimates and decent interdependence between parameters, we concluded that the SgrS growth knob model is fully identifiable. The results of this analysis strengthened the reliability of the predictions made using the SgrS growth knob model.Rotem, E., Loinger, A., Ronin, I., Levin-Reisman, I., Gabay, C., Shoresh, N., … Balaban, N. Q. (2010). Regulation of phenotypic variability by a threshold-based mechanism underlies bacterial persistence. Proceedings of the National Academy of Sciences of the United States of America. https://doi.org/10.1073/pnas.1004333107
Fei, J., Singh, D., Zhang, Q., Park, S., Balasubramanian, D., Golding, I., … Ha, T. (2015). Determination of in vivo target search kinetics of regulatory noncoding RNA. Science. https://doi.org/10.1126/science.1258849
Kuzmanovska, I. (2012). Markov chain Monte Carlo methods in biological mechanistic models. ETH Zurich. https://doi.org/10.3929/ethz-a-007343293