Engineering Basal Cognition: Minimal Genetic Circuits for Habituation, Sensitization, and Massed-Spaced Learning
Sensitizationalso called facilitationis another fundamental form of nonassociative learning that stands in direct contrast to habituation. While habituation involves a decrease in response after repeated exposure to a stimulus, sensitization is characterized by an amplification of the response after repeated or intense stimulation.
This heightened responsiveness serves as an essential adaptive mechanism across a wide range of species, enabling organisms to prioritize and react more vigorously to potentially harmful or biologically significant stimuli. Sensitization has been extensively studied in the context of defensive behaviors and nociceptive responses, where it enhances survival by promoting rapid responses to environmental threats. Beyond behavioral responses, sensitization manifests itself at multiple levels of biological organization, from whole-organism reflexes to synaptic plasticity and cellular signaling pathways.
The molecular mechanisms that underlie sensitization are generally based on positive feedback loops or cascade amplification systems, which progressively strengthen in response to repeated stimulation. In this work, a simple synthetic design that recapitulates this behavior is presented, drawing upon elements from our earlier design for habituation. The key to this system lies again in leveraging the molecular memory provided by repressor I, which generates a signal that accumulates in the presence of input and decays slowly during the relaxation phase; this dynamic enables the circuit to effectively “remember” prior stimulations and respond more strongly upon repeated exposure. It is also worth noting that, in the context of synaptic plasticity, sensitization is often achieved through a basic motif derived from the habituation mechanism, with the addition of an extra interneuron.
Building on the basic logic of sensitization illustrated in Figure a, a functional genetic circuit can be derived. This proposed circuit (Figure b) largely reuses the gene regulatory interactions from the previous habituation design but incorporates a few key modifications to establish a positive feedback loop.
The circuit creates a positive feedback loop on receptor X (LuxR-SsrA) through a double repression cascade. In the presence of input x, the X–x complex drives expression of repressor I (LacI), which in turn represses expression of R (TetR-SsrA). Since R itself represses the production of X, the net effect is that X indirectly promotes its own expression and amplification. This design decouples the feedback strength through intermediate regulators, allowing tunable and graded amplification that depends on the stimulation history. By using this indirect mechanism, the system avoids some limitations inherent to a direct self-activation loop (see Supporting Information Subsection S4.2).
The output G (GFP-SsrA), which is expressed from a promoter directly activated by the X–x complex, is consequently also amplified. Thus, the production of G is dependent upon both the presence of the external input and the time-dependent concentration of X, ensuring input-gated activation and a sensitized response that reflects the system’s history of stimulation. An alternative design based on a coherent feed-forward loop (C-FFL), which offers similar performance and complexity, is discussed in the Supporting Information (Subsection S4.3).
The key to the memory effect is that only the repressor I has a slower degradation–dilution rate (γ). Its continued presence represses R, which in turn relieves repression of the receptor X promoter. The fast degradation rate (λ) of both X and R ensures their rapid turnover, with their concentrations rapidly approaching a quasi-steady-state set by the slower dynamics of I, enabling time scale separation and effective memory encoding.
The equations governing the system are as follows: dXdt=μθ−(R)−λX,dIdt=αθ+(Xx)−γI,dRdt=ρθ−(I)−λR,dGdt=βθ+(Xx)−λG 9 where α, β, μ, ρ represent independently tunable maximal expression rates, γ is the basal rate at which proteins dilute–degrade, and λ is the degradation rate of short-lived (tagged) proteins. As before, the constraint λ ≫ γ is assumed, creating a difference of time scales. This ensures that untagged proteins degrade rapidly, while other proteins accumulate within the cell over successive pulses.
An example of a time-series that displays sensitization is shown in Figure d. As in the habituation example, the memory component I accumulates over timebut here, this leads to a decrease in repressor R, which otherwise inhibits production of X, effectively boosting receptor levels in proportion. This double inhibition loop effectively increases sensitivity to input x, yielding progressively larger peaks in output G.
Figure c shows the parameter space (α, γ) for the sensitization model, where the fold change from eq is used to display a well-defined boundary of sensitization responses (blue domain). The decrease in the fold change as the production rate of the memory molecule (α) increases is due to the faster accumulation of memory I, which more quickly saturates its corresponding transfer function. When α is high, the system further approaches this saturation limit during the first peak response, thereby limiting any further increase in the levels of the input receptor and the output protein (see Supporting Information Subsection S4.1 for details).
Cognition is not limited to specific kinds of response, and a relevant question is how we can combine or modify our minimal motifs in such a way that multiple learning responses can be at work. This occurs with some simple neural circuits that can encode both increased and decreased responsiveness. This is the case of classic experiments in Aplysia californica, where a slight touch to the siphon triggers a gill-withdrawal reflex. , After a strong stimulus such as a tail shock, this response becomes sensitized, with even light touches causing a stronger reaction. However, if light touch is repeated without reinforcement, the reflex gradually habituates, weakening over time. While sensitization involves enhanced neurotransmitter release, habituation results from reduced synaptic activity.
In this section, we briefly illustrate the possibility of combining sensitization and habituation responses within a single circuit to produce a hybrid response. This results in an output where an initial period of sensitization is followed by habituation.
Although one might expect this construct to require some nontrivial combination of the previous motifs, surprisingly, this more complex behavior does not require a fundamentally new design (Figure a). It can be achieved simply by reintroducing a single repressor interaction from the habituation circuit, the hybrid promoter from Figure b, into the sensitization design from Figure b. In this hybrid circuit, output G (GFP-SsrA) expression depends on both the presence of the input-bound receptor X–x and the absence of repressor I. The behavioral transition occurs when the inhibitory effect of I on output expression dominates over the activation pathway mediated by X–x.
The number of pulses required to transition from sensitization to habituation is tuned by the relative sensitivity to I for the two promoters regulated by it. If the hybrid promoter requires a higher repressor concentration, a clearer separation between the two phases can be observed. This sensitivity can be experimentally engineered through mutagenesis or predicted in silico.
An alternative implementation, tweaking a C-FFL sensitization motif, is also possible by using two distinct sources for repressor I, with different dynamics: a tagged version, indirectly repressed by the input complex for sensitization; and an untagged version, directly expressed by the same complex for habituation (see Supporting Information S5.3).
The equations governing the system are as follows: dXdt=μθ−(R)−λX,dIdt=αθ+(Xx)−γI,dRdt=ρθ−(I)−λR,dGdt=βθ+(Xx)θK‐(I)−λG 10 where α, β, μ, ρ represent independently tunable maximal expression rates, γ is the basal rate at which proteins dilute–degrade, and λ is the degradation rate of short-lived (tagged) proteins.
The function θK−(I) represents the repressor curve for the output. It is analogous to the repressor transfer function presented in eq , but with an increased half activation constant K, θK−(I)≔11+(IK)2 11 This increase in K 1/2 shifts the system’s dynamic response to a longer characteristic time scale. Consequently, the point at which the system’s behavior shifts from sensitization to habituation occurs more slowly.
Figure d illustrates an example time-series in which the memory component I accumulates progressively over time, as observed in earlier cases. As I builds up, receptor X increases, leading to an amplified response and successively larger peaks of output G. However, this positive feedback eventually saturates, even as I continues to accumulate. Beyond a critical threshold, further increases in I have a stronger inhibitory effect on G than the activation mediated by X, leading to a progressive decline in the peak amplitude of output G.
Figure b,c shows that the parameter regions for sensitization and habituation in the hybrid circuitdefined by a doubling or halving of the output peakresemble those of the individual circuits (Figures c and c). Although these regions show partial overlap, the two behaviors cannot be simultaneously maximized. This stems from their divergent parameter dependencies: while both behaviors require a low memory degradation–dilution rate γ, habituation strengthens monotonically with the maximal expression rate α, whereas sensitization weakens beyond an optimal α value. This inherent trade-off is a direct consequence of sharing core circuit components for both motifs within a single, shared pathway.
Our last example is the so-called massed–spaced learning (MSL). It refers to a cognitive strategy in which learning sessions are distributed over time (spaced learning) rather than concentrated over a short period (massed learning). This approach improves memory retention and recall by allowing time for consolidation between learning episodes. Massed–spaced learning could involve synaptic plasticity mechanisms at the single-cell level, where neurons strengthen or weaken their connections based on repeated but temporally spaced stimulation. Key processes such as long-term potentiation and synaptic tagging can be the basis for this effect, allowing individual neurons to encode and maintain learned information more effectively when spaced intervals optimize molecular and structural changes. It can be characterized on multiple scales, from synaptic changes to individual behavior, and manifests itself at both the behavioral and molecular levels.
Although MSL is traditionally considered a neural process, a recent study suggests that similar behaviors may also be present in aneural systems, including single cells within a given tissue. Their findings indicate thatunder controlled conditionsthese non-neural cells seem to retain information better when they are exposed to spaced intervals rather than all at once. The authors used an engineered non-neuronal reporter cell line capable of exhibiting spacing-dependent responses, offering not only increased experimental throughput to model memory formation, but also a platform to study cellular cognition outside the nervous system.
The logic architecture of MSL can be captured by a cascade of positive effects, as sketched in Figure a. Once again, a memory component is a requirement for learning capacity, but now the output response can be decoupled from the immediate presence of input events. This decoupling arises because the input stimulus does not directly drive output expression; instead, it triggers the synthesis of an intermediate transcriptional activator that persists transiently after stimulus withdrawal. Consequently, sustained output gene expression can occur as long as sufficient concentrations of this inducer remain, even in the absence of ongoing input. This behavior differs fundamentally from all previously described motifs, in which output synthesis occurs only while the input signal is present, and intermediate components modulate the amplitude of the response rather than directly drive output production.
The MSL phenomenon can now be observed (under the right conditions) by dividing a sustained stimulus into a series of shorter stimuli separated by a delay Δt, in such a way that the cell can accumulate inducer molecules over time, leading to a higher overall response. However, as will be shown below, this enhancement works only when the intervals between stimuli are shorter than the characteristic decay time of the intermediate inducer. If the delay between stimuli exceeds this critical duration, the concentration of the inducer will diminish substantially before the arrival of the subsequent stimulus, thus negating the cumulative effect.
A minimal genetic implementation of this circuit, as shown in Figure b, consists of a linear feed-forward chain where a receptor protein X detects an external signal molecule x, leading to the production of an inducer A, which in turn controls the expression of the output protein G. Specifically, we suggest the input signal C6–HSL is integrated by accumulating an inducer, AraC, resulting in prolonged GFP expression that lingers after the input vanishes.
The equations governing the system are now: dXdt=μ−γX,dAdt=αθ+(Xx)−γA,dGdt=βθ+(A)−γG 12 where α, β, μ are different promoter strengths that can be tuned separately, and γ is the rate at which proteins dilute–degrade.
The dynamical response of the circuit is summarized in the time-series in Figure c. For a single, continuous input pulse (left column), the intermediate activator A accumulates rapidly during the stimulus period, followed by an exponential decay afterward. This transient accumulation produces a single maximum peak in the output, which quantifies the response efficiency. In contrast, when the same total input duration is divided into several evenly spaced pulses (right column), the concentration of A accumulates incrementally and decays only partially during the interpulse intervals. This pattern of spaced reinforcement results in a higher maximum peak amplitude of the output compared to the massed case, demonstrating the expected enhancement in response. The circuit is therefore capable of distinguishing between continuous (massed) and temporally spaced input patterns, analogous to the way neural systems differentiate between massed learning and learning reinforced through repeated, spaced exposures over time.
A parameter space illustrating the relative efficiency of distributing the input into N evenly spaced pulses, separated by interpulse intervals Δt, is shown in Figure d.
The log2-transformed fold change in the peak response relative to the massed input case is used to quantify the enhancement due to input spacing: FC≔log2maxG(t)|N,ΔtmaxG(t)|N=1 13 where the case N = 1 corresponds to massed input, and no interpulse delay is therefore applicable. Positive values of FC indicate an increased peak response under spaced stimulation, with the magnitude reflecting the extent of enhancement compared to the massed case.
As expected, a region exists in which dividing and spacing the input pulses yields a positive gain; however, both parameters are interdependent and must be balanced.
For very short interpulse delays (Δt → 0), the difference becomes negligiblein this limit, the signal effectively behaves as a continuous (massed) input, provided N remains finite and physiologically realistic.