Decision making circuitry in bacteria

Bacteria make decisions one of which is to swim towards food (or away from repellents ) using molecular signaling networks inside its “body” — a single cell.

There is no “brain” in bacteria in the sense we use the word “brain” for multicellular life — since bacteria are single celled life forms. However, inside in these single cell life forms are molecular signaling networks working at different time scales that drive decision making required for its survival, most of which have been conserved in multicellular life forms too.

  • When bacteria (E. coli) is in an environment with abundant nutrients, it continues to divide and does not try to move. [1]
  • When conditions becomes worse, they make a decision to grow several nanometer sized motors attached to propellers (flagella) which allow it to swim. It also generates a navigation system that tells it where to go in search of food avoiding repellents along the way.
  • They can sense molecule gradients (e.g. food molecule) as small as one molecular per micron in a background of 1000 molecules per cell volume. Their small “body” length of one micron prevent this high level of sensitivity to food detection using antenna on the ends of the cell (spatial gradient sensing). The random fluctuations (Poisson) of the background signal (sqrt(1000) ~ 30) would mask the one molecule per micron gradient that they can detect. These organisms instead sample the gradient over time and integrate it, as opposed to spatially integrating the gradient. [2]

Food search and adaptation behavior in a fluid environment:

  • “Search decision” when no food is detected. They perform a random walk scan which is composed of straight line motion (runs) lasting about a second interleaved with tumbles lasting 100 msec that randomly change direction. Figure 1
  • “Go to food decision” when detecting a food molecule gradient. Within 100 msecs of detecting a food molecule gradient, their tumbling frequency will plummet, making them essentially swim (just runs — no tumbles) along the gradient.
  • “Adapt decision” after being presenting food environment for a while (several minutes). Once they are in the food environment for a few minutes, they start to tumble again and continue their exploration. This adaptation behavior allows for the bacteria to maximize it scan of its environment, conferring it with some benefits (perhaps not being stuck in local optimum. Contrast this however with the sedentary behavior before the production of motors mentioned earlier. There is however a hedging bet decision too in sedantary life mentioned below).

Molecular signaling mechanisms driving each decision above

  • “Search decision” (when no food is detected):
  • A bacterial cell (E. coli) has several thousand sensors on embedded in the “body surface” (cell membrane) that can sense molecules of interest (e.g. food).
  • These sensors are attached on the inside of the cell to a “toggling molecule” that rapidly toggles in the absence of food between active and inactive states in microseconds.
  • When it is in an active state, the “toggling molecule” can activate a specific “messenger molecule”.
  • These activated “messenger molecules” can diffuse into the cell and go and bind to a motor and change the direction of rotation of the propeller.
  • This in turn causes the bacteria to tumble. The activated “messenger molecules” are constantly removed by another “scavenging molecule”.
  • So in steady state (or idle state), when no food is detected, the “toggling molecule” toggles continuously, effectively dispatching “messenger molecule” to the motors causing the bacteria to tumble.
  • The opposing action of the “scavenging molecule” and the “messenger molecule” coupled with the continuous toggling of “toggling molecule” in turn activating the “messenger molecule” leads to a steady state tumbling frequency.
  • This simple mechanism of interactions between “toggling molecule”, “messenger molecule”, and “scavenging molecule” implements the search decision which is effectively a random walk scan of the environment.
  • “Go to food decision” (when detecting a food molecule gradient):
  • When a food molecule binds to a sensor on a cell’s body, the binding suppresses the toggling of the “toggling molecule” between active and inactive states. This in turn reduces the number of active “messenger molecules” that can bind to motor, and hence the tumbling plummets and “runs” dominate effectively enabling the bacteria to progress towards the food. The average diffusion time of a “messenger molecule” to a motor is 100 msec, so one can observe tumbling frequency plummet within 100 msec of sensing food.
  • “Adapt decision” (after being presenting food environment for several minutes).
  • The sensors are attached on the inside of the cell not only to a “toggling molecule”, but also to “auxiliary activation” switches that when turned on can restore the activity of the “toggling switch” despite the presence of food attached to the sensors.
  • The “auxiliary switches” on sensors can be activated by an “auxialiary switch activator” molecule and turned off by an “auxiliary switch deactivator” molecule.
  • These two opposing molecules keep turning the “auxiliary switch” on and off even when there is no food present (i.e. during random walk searching for food).
  • This apparenly wasteful action has an important function. It allows the cell to adapt as explained below.
  • The toggling molecule not only activates the “messenger molecule” as mentioned earlier, it also activates the “auxiliary switch deactivator”.
  • So when food is found and the toggling of the toggling molecule is suppressed, it also in turn suppresses the activation of “auxiliary switch deactivator”. This allows “the auxiliary switch activator” to dominate and turn on the “auxiliary switches”, and this in turn restores the activity of the toggling switch, which then dispatches “messenger molecules” to the motor to cause tumbling.
  • The feedback loop of the “auxiliary switch” signalling pathway is slower than than “messenger molecule” pathway — it operates in the minutes scale as opposed to the hundreds of milliseconds scale. So adaptation kicks in much later.

Other examples of decision making in bacteria

  • Assembly of motor from parts (Figure 2)
  • When conditions are not favorable, as mentioned earlier, bacteria decide to grow 50 nm sized motors made up 30 types of parts (proteins) than can propel it forward thirty times it body length/sec (30 microns/sec)
  • The motor is assembled in stages like in an assembly line, where the part made in each stage diffuses into the part already assembled.
  • This orchestrated assembly is made possible by a First in First Out(FIFO) manufacturing by a network motif as shown in Figure 3 below.
  • Two molecules X and Y play a role in this assembly — assembly can happen either in the presence of X or Y. X not only influenes the production of parts, it alone is drives the production of Y. To begin with, as X increases in concentration, manufacturing of part 1 of motor commences — this concentration of X is less than that required for part 2 to begin , and so on. So part 1 manufacturing begins and as X increases part 2 manufacturing starts, and so on. Since the production of Y depends on X, Y starts building up as X builds. So in essence the assembly line of parts get turned on one after the other as shown in Figure 3.
  • The turning off manufacturing in FIFO manner is made possible by the dependencies of the manufacturing of parts on the concentration Y. Once X is turned off, the concentration levels of Y starts to drop. The manufacturing of part 1 can proceed with just Y ( it’s an OR dependency — just X or needs to be present) but it requires a higher concentration of Y than part 2, and part 2 requires more of Y than part 3, and so on. Given this dependency, once X stops, and Y starts to fall, manufacturing of part 1 stops, followed by part 2. This form of dependency of the parts on the concentrations of X and Y enables a FIFO manufacturing pipeline.
  • Stochastic switching of cell fate
  • Bacteria also choose their fate stochastically without apparent regard to environment or history. [3]
  • This form of stochastic control appears to have some advantages. For instance, populations of bacteria have subpopulations that have entered a non growing state even in the presence of food. By this strategy, it appears as though the population as a whole is hedging its bets against the possibility of an adverse condition such as the application of antibiotics, which can only kill active bacteria, not the “non growing ones”

In summary, bacteria have a wide range mechanisms to perform decision making, and these mechanisms and variations/improvements of them are found in multicellular life forms including humans.



Figure 1. The chemotaxis system in E. coli A, CheA histidine protein kinase (“the toggle molecule”); W, CheW linker protein between receptors and CheA; Y, the response regulator CheY (“the messenger molecule”); Z (“the scavenging molecule”); B, the response regulator CheB (“the auxillary switch deactivator”); R, the methyltransferase CheR (“the auxillary switch activator”); P, phosphoryl group. The flagella motor is shown at the right of the Figure.
Figure 2. The flagellar motor of E.coli and its assembly steps.
Figure 3. Schematic of the parts (proteins) manufacturing circuit that implements a First In First Out (FIFO) assembly line. The gene transcription for each part Z1, Z2, Z3, require the presence of either molecules X or Y( OR gate). However the dependancy on the concentatraions are in reverse relationships. Specifically, the dependancy on X is such that part 1 needs less of X than part 2, which in turn needs less of X than part 3 (S1 < S2 < S3). However, the dependancy of Y is just the opposite (E1 > E2 > E3). As explained in the main text, this reverse ordering enables implementation of a FIFO pipeline. If the dependency on the concentration on Y was the same as on X that would be a stack ( Last In First Out). So this network motif can implement a stack or a queue if the concentration dependancy on X and Y was tweaked. Also, the the dependency on X and Y are not in absolute amounts — there is too much variation in production levels of molecules to depend on absolute levels. For example, if the production of X is varied in absolute concentrations the relationship of turning on sequence is still not altered since it is only driven by the inequality like S1 < S2 < S3. Also the turning on and turning off of genes expression is dependant on the molecular structure of the attaching and substrate molecule ( disassociation and association constants ) which is largely invariant. So this assembly process is quite robust to the absolute levels of production of X and Y which indeed is the case in cells — production levels vary from cell to cell. This basic fact is a design methodology that has been conserved in multicellular life forms too — robustness of design to absolute levels of production of molecules. [1]


  1. An Introduction to Systems Biology: Design Principles of Biological Circuits Uri Alon,2006
  2. Adaptation and control circuits in bacterial chemotaxis 2010.
  3. Stochasticity and cell fate,2008.

Originally published at

Machine learning practitioner

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store