Skip to content

Eavesdropping on bacteria with CellEngine

No one is surprised to read that individuals coordinate their behavior and act as groups. We see collective behavior everywhere: from family gatherings, to corporate workplaces, and on the playing field. We also know that other animals act collectively. Birds flock, fish school, and wolf packs hunt. What might surprise us, however, is how ubiquitous the phenomenon is. Remarkably, coordinated, collective behavior can be seen in organisms as simple as bacteria, single-celled creatures without even a nucleus, much less a nervous system. Because they are so simple, bacteria can serve as a model organism for studying collective behavior. And because each individual is a single cell, CellEngine can be an important tool for analyzing that behavior.

Sensing a Quorum

One mechanism that bacteria use to coordinate their behavior is Quorum sensing. QS alters the expression of key genes depending on the density of the local bacterial population. For example, the amount of an enzyme that an isolated bacterial cell can produce on its own may be ineffective. But when many bacterial cells are crowded together, they can, by pooling their resources, create enough of that enzyme to alter their environment. Quorum sensing can control expression of the gene for the enzyme. It supresses expression when cells are alone and activates expression when they are populous.

Quorum sensing relies on small molecules known as signals. Within each cell, signal synthase proteins produce signal at a relatively steady rate. The signal diffuses (or is transported) across the cell membrane into the environment. Cells also contain signal receptor proteins to which signal molecules can bind. The complexes formed by signals bound to receptors serve as transcription factors controlling the expression of other bacterial genes. When the bacterial population is sparse, there is a limited amount of signal to form these transcription factors. In dense environments, on the other hand, signal is plentiful and an ample supply of transcription factors is available. The gene regulatory networks behind quorum sensing frequently include feedback loops; signal-receptor transcription factors not only regulate behavior such as enzyme production, they can also increase expression of the genes responsible for the synthase and receptor proteins themselves. In this way, quorum sensing creates positive feedback that significantly amplifies its effect.

When it was first recognized, quorum sensing was thought to be relatively rare among bacteria1. Since its initial discovery, however, many species have been found to rely on quorum sensing for a diverse set of behaviors. In fact, several species contain multiple, parallel quorum sensing networks in their genomes2. This observation suggests that quorum sensing might do more than simply estimate population density. A single quorum sensing network would be sufficient for that function, and we might expect evolution to dispense with any redundant gene regulatory networks.

Quorum Sensing in Pseudomonas aeruginosa

One bacterial species with multiple quorum sensing networks is the opportunistic pathogen Pseudomonas aeruginosa. Its genome includes two particularly well-studied networks: the las network the and the rhl network. To investigate potential fitness benefits of multiple QS systems, my colleagues and I compared those two systems as colonies of P. aeruginosa PAO1 strain grew from low to high population densities over the course of about 18 hours3. As expected, we observed that both systems behaved similarly at low densities and high densities. The behavior at intermediate densities, however, was somewhat surprising. For a transient period while the colonies were growing, the cells segregated into to two distinct subpopulations: in some cells the las network was more active than the rhl network, while in other cells the opposite was true. These different populations formed even though growth in this case was strictly clonal with all cells in both populations having the same genome and experiencing the same environment.

Evidence for this behavior can be found in flow cytometry data, an example of which is shown in the figure below. Here we are looking at the expression level of the receptor genes, lasR and rhlR respectively for 50,000 cells. To collect this data, we used chromosomally integrated fluorescent gene reporters: GFP for lasR and mCherry for rhlR. The figure shows results nine hours after the cultures were inoculated, about half of the time needed to reach maximum population density. One population preferentially expresses lasR while the other preferentially expresses rhlR.

The transient nature of this phenomenon can be seen by looking the correlation coefficient of the expression levels across time. The following table lists those measurements for both lasR/rhlR and lasR/lasB. (Somewhat confusingly, lasB is a gene known to be controlled by both the las and rhl systems4; it is not a direct part of the las system proper.) In both cases the negative correlation, indicating multiple subpopulations, peaks around 9 hours.

Time Point 𝜌 (lasR/rhlR) 𝜌 (lasR/lasB)
0 hrs 0.17 0.17
3 hrs 0.11 0.04
6 hrs 0.06 0.03
9 hrs βˆ’0.38 βˆ’0.39
12 hrs βˆ’0.26 βˆ’0.31
15 hrs βˆ’0.12 βˆ’0.27
18 hrs 0.09 βˆ’0.13

Competition for a Constrained Resource

Although we do not propose a specific mechanism for the transient, negative correlations, we hypothesize that it may be due to resource constraints. In particular, both the las and rhl systems may need a common resource to activate expression of their respective genes. During the time period in which multiple populations appear, that resource is in short supply. When one gene expression pathway makes use of this resource, less of the resource is available for the second pathway.

Michaelis-Menten kinetics5 provides a simple model for such a system. Although initially developed for analyzing enzymes, the approach can also describe gene transcription, signal transduction, and other biological processes. For this model, we consider simple, second order reactions. We consider a single ligand (the constrained resource) that reversibly binds with two receptors (representing the two QS networks).

$$ \begin{align} L + R_1 & \rightleftharpoons L R_1 \\ L + R_2 & \rightleftharpoons L R_2 \nonumber \end{align} $$

In this model the rates are $k^+_1$ and $k^-_1$, and $k^+_2$ and $k^-_2$ respectively, resulting in disassociation constants $Kd_1 = k^-_1 / k^+_1$ and $Kd_2 = k^-_2 / k^+_2$. We consider expression of the two genes in a gene pair as requiring the bound complexes $LR_1$ and $LR_2$ respectively. The actual regulatory network is likely more complicated, as it may involve multiple enzymes, receptors, or transcription factors; however, the minimal model captures the essential behavior of the system.

Evaluating the Hypothesis

To evaluate this hypothesis we simulate the two-equation system in MATLAB using Gillepsie's algorithm6. The implementation allows us to vary the quantity of ligand $L$ relative to the receptors and as well as the disassociation constants of each reaction $Kd_1$ and $Kd_2$. For each set of values, we simulate 50,000 cells and observe the correlation of the two bound complexes $LR_1$ and $LR_2$.

for Step = 1 : (Options.NumSteps - 1)
  Propensities = [
    RateConstants(1) * Counts(Step, 1) * Counts(Step, 2)
    RateConstants(2) * Counts(Step, 4)
    RateConstants(3) * Counts(Step, 1) * Counts(Step, 3)
    RateConstants(4) * Counts(Step, 5)
    RateConstants(5) * Counts(Step, 1) * Counts(Step, 6)
    RateConstants(6) * Counts(Step, 7)
  ];
  TotalPropensity = sum(Propensities);
  RandomNumbers = rand(1, 2);
  NextTime = -log(RandomNumbers(1)) / TotalPropensity;
  NextReaction = find( ...
    (cumsum(Propensities) >= RandomNumbers(2) * TotalPropensity), ...
    1, 'first' ...
  );

  Time(Step + 1) = Time(Step) + NextTime;
  Counts(Step + 1, :) = Counts(Step, :) + Reactions(NextReaction, :);
end

Comparing the Results

To check the model's applicability to our experimental observations, we generate synthetic flow cytometry data from its output. The following figure compares this synthetic data with our experimental observations. When resource availability is low, the model reproduces the negative correlation observed at the intermediate time point. At high resource availabilty the negative correlation disappears, and the model corresponds to experimental observations at later time points.

Conclusions

This post summarizes an application of CellEngine to microbiology research. That research explores the activation of quorum sensing gene networks in a clonal population of bacterial cells. Even though the cells have the same genome and experience the same environment, CellEngine analysis demonstrates that the colony segregates into two distinct subpopulations during one phase of its growth. We also generate sythetic data from a model of our hypothesis and use CellEngine to visually demonstrate the plausibility of the hypothesis.

About the author: Stephen Thomas is an engineer on the CellEngine team focusing on user interface development. He holds a Ph.D. in Quantitative BioSciences from the Georgia Institute of Technology.

  1. Bassler BL, Losick R. Bacterially speaking. Cell. 2006 Apr 21;125(2):237-46. doi: 10.1016/j.cell.2006.04.001. ↩

  2. Papenfort K, Bassler BL. Quorum sensing signal-response systems in Gram-negative bacteria. Nature Reviews Microbiology. 2016 Aug 11;14(9):576-88. doi: 10.1038/nrmicro.2016.89. ↩

  3. Jayakumar P, Thomas SA, Brown SP, KΓΌmmerli R. Collective decision-making in Pseudomonas aeruginosa involves transient segregation of quorum-sensing activities across cells. Current Biology. 2022 Dec 19;32(24):5250-5261.e6. doi: 10.1016/j.cub.2022.10.052. ↩

  4. Pearson JP, Pesci EC, Iglewski BH. Roles of Pseudomonas aeruginosa las and rhl quorum-sensing systems in control of elastase and rhamnolipid biosynthesis genes. Journal of Bacteriology. 1997 Sep;179(18):5756-67. doi: 10.1128/jb.179.18.5756-5767.1997. ↩

  5. Srinivasan B. "A guide to the Michaelis-Menten equation: steady state and beyond." The FEBS Journal. 2021. doi: 10.1111/febs.16124. ↩

  6. Gillespie D. "Exact Stochastic Simulation of Coupled Chemical Reactions." The Journal of Physical Chemistry. 1977;81(25):2340-2361. doi:10.1021/j100540a008. ↩