Engineering Basal Cognition: Minimal Genetic Circuits for Habituation, Sensitization, and Massed-Spaced Learning

Cognition is often associated with complex brains, yet many forms of learningsuch as habituation, sensitization, and even spacing effectshave been observed in single cells and aneural organisms. These simple cognitive abilities, despite their cost, offer evolutionary advantages by allowing organisms to reduce environmental uncertainty and improve survival. Recent studies have confirmed early claims of learning-like behavior in protists and slime molds, pointing to the presence of basal cognitive functions long before the emergence of nervous systems. In this work, we adopt a synthetic biology approach to explore how minimal genetic circuits can implement nonassociative learning in unicellular systems. Building on theoretical models and using well-characterized regulatory elements, we design and simulate synthetic circuits capable of reproducing habituation, sensitization, and the massed–spaced learning effect. Our designs incorporate activators, repressors, fluorescent reporters, and quorum-sensing molecules, offering a platform for experimental validation. By examining the structural and dynamical constraints of these circuits, we highlight the distinct temporal dynamics of gene-based learning systems compared to neural counterparts and provide insights into the evolutionary and engineering challenges of building synthetic cognitive behavior at the cellular level.

Life on Earth has evolved in multiple directions, with major evolutionary events defining the rise of novelties, such as the transition from unicellular to multicellular or the emergence of language. , A general trait shared by most of these transitions is the emergence of new types of agents capable of dealing with environmental uncertainties in novel ways. Among others, the evolution of neural agents represents a revolutionary path toward neural networks and brains. The rise of cognitive structures and brains largely begins with the Cambrian explosion event, where we can see the rapid evolution of animals equipped with mobile parts and sensors. Using a special class of cells, the neuron, it was possible to develop mechanisms to store and process information reliably. In this context, learning might have played a crucial role in the development of complex brains.

Cognitive structures are costly, from sensors and actuators to the whole brain. How can evolution favor their emergence? The answer is that mastering time, i.e., storing past information that can be used to predict the environment, can have a high return. In other words, reducing uncertainty has a high pay-off. Not surprisingly, memory is a widespread feature of most multicellular life, and several kinds of learning mechanisms have been described. These include two well-known processes, namely habituation and sensitization, which have been known for thousands of years as well as associative learning. However, these features are not limited to multicellular systems and were identified by several authors regarding the behavioral responses of individual cells. This is the case of Stentor (Figure a), a single-cell protozoan , which was shown to exhibit an enhanced response to subsequent milder stimuli, that is, habituation. This indicated that it was learning to associate the intense stimulus with a potential danger, resulting in increased sensitivity. Similarly, Stentor was also shown to exhibit a reduced response under repeated exposure to the same stimulus, indicating that it was learning to recognize the stimulus as nonthreatening or irrelevant. In addition, sensitization, when an organism becomes more sensitive to a stimulus after experiencing an intense or aversive event, was also observed.

Although Jennings’ work was questioned due to reproducibility issues, recent studies confirmed (and expanded) his key results, uncovering a hierarchy of decision-making actions. − Both habituation and sensitization, along with other forms of learning, have also been found in Physarum polycephalum (Figure b). This aneural organism is a giant multinucleated cell that can extend up to hundreds of square centimeters. It has been shown to exhibit complex, optimal self-organized patterns and solve a wide range of tasks that involve complex computational decisions. −

Basal cognition encompasses the fundamental processes and mechanisms that enable organisms to detect certain environmental states and respond appropriately to ensure survival (such as finding food and avoiding danger) and reproduction long before the evolution of nervous systems. Modern cells have complex molecular networks that can participate in cognitive tasks, from detecting and responding to chemical cues to actively exploring their worlds in space and time. An important aspect of their behavioral repertoires is the presence of memory and learning characteristics. They are, in fact, strong prerequisites to build a behavior. In recent years, renewed interest has emerged in this area, including dedicated efforts to operationally reframe cognition.

Theoretical and computational studies on these simple forms of learning have recently been developed, searching for potential circuits capable of implementing them within cells, particularly in terms of habituation. , There is an alternative approach to this problem from an engineering perspective. In general, while much understanding of major evolutionary transitions has been gained through molecular phylogenetics or comparative analysis and paleobiological data, we can also address the problem by recreating these evolutionary events and their precursors using (among other approaches) synthetic biology. In this context, the design of genetic circuits and the potential of tissue bioengineering can help understand the constraints associated with biological complexity. This includes morphology, − multicellularity, , ecology, , biorobots or collective intelligence, among other problems.

Previous studies have addressed the problem of classical conditioning (associative learning) circuits based on transcriptional network systems, including several proposals for candidate molecular circuits. − In this paper, we take the “synthetic” approach to consider how simple genetic circuits could reliably implement nonassociative learning, thus involving a single type of stimulus. This will include the two previous case studies (habituation and sensitization) and the so-called massed–spaced learning effect. The latter has traditionally been discussed within the psychology literature and involves scenarios in which long-term memory is enhanced when learning events are spaced apart in time, rather than occurring in immediate succession. However, recent studies have shown that it can also be present in single human cells, suggesting that this learning effect might also operate on a small scale within whole bodies.

Our implementations rely on well-characterized genetic parts and regulatory mechanisms, making them amenable to experimental realization. The designed circuits proposed here will include repressors, fluorescent reporters, and quorum-sensing molecules as key components, allowing external control of the input signal and measurement of the average output at the population level. While the specific examples are tailored for E. coli , the underlying regulatory motifs are abstract, modular designs rooted in transcriptional logic and are not inherently limited to bacterial systems. By following our synthetic approach, there is a marked difference between the gene networks that implement learning and their synaptic counterparts: a much slower time scale. This difference will be relevant to our discussion on the role that learning plays in single-cell organisms and the constraints imposed on engineering designs.

Habituation represents one of the simplest and most ubiquitous forms of nonassociative learning, characterized by a gradual decrease in the behavioral or physiological response after repeated exposure to a stimulus that is perceived to be neither harmful nor beneficial. This phenomenon is not merely passive fatigue or sensory adaptation, but an active process. It has been extensively studied across various speciesfrom Aplysia to humansdemonstrating its evolutionary conservation and functional importance in adaptive behavior.

This fundamental mechanism of behavioral adaptation allows organisms to efficiently filter out irrelevant or nonthreatening stimuli from their environment, such as background noise or persistent visual cues. By reducing responses to familiar inputs, habituation enables the organism to conserve cognitive and energetic resources, which can then be redirected toward detecting and responding to novel or potentially significant environmental changes. In this way, habituation plays a crucial role in attentional modulation and information processing, helping organisms prioritize survival-relevant stimuli.

At its core, habituation requires a responsive system capable of detecting external stimuli and modulating its sensitivity over time. This modulation typically involves negative feedback mechanisms that accumulate with repeated stimulation. For example, repeated activation of postsynaptic receptors may lead to internalization of these receptors or alterations in second-messenger signaling pathways, effectively raising the threshold required to elicit a response. Such mechanisms enable the system to adaptively ignore persistent signals while remaining sensitive to new or changing inputs.

Figure a sketches the logic of our proposed minimal circuit, and in Figure b a genetic design is shown as a realization of an incoherent feed-forward loop (I-FFL), a motif previously discussed in the literature. ,, Alternative designs based on negative feedback topologies for the receptor are discussed in the Supporting Information (Subsections S3.2 and S3.3), where they are shown to perform less optimally.

The proposed synthetic circuit senses and reacts to an external input signal x, such as the quorum-sensing molecule C6–HSL, which diffuses into the cell and binds to a constitutively expressed transcriptional regulator LuxR-SsrA, denoted X. The resulting ligand-bound complex X–x activates transcription of two key components: a memory I and an output G.

The memory component I, the repressor protein LacI in the example, is expressed under an inducible promoter activated by X–x, and its concentration accumulates over successive input pulses, serving as a biochemical integrator of past input activity. Output G, represented for simplicity as the fluorescent reporter GFP-SsrA, is expressed under a hybrid promoter that requires the presence of complex X–x for activation while simultaneously being repressed by I, which approximates N-IMPLY logic (x ∧ ¬I). However, rather than enforcing a sharp binary switch, intermediate concentrations of I lead to partial suppression of promoter activity, effectively reducing the output production rate.

All proteins except memory I contain a degradation tag that enhances their proteolytic turnover, resulting in a higher degradation–dilution rate. This difference in stability between sets of proteins establishes a difference of time scales, ideally allowing only untagged proteins to persist between input pulses.

The equations governing the system are as follows: dXdt=μ−λX,dIdt=αθ+(Xx)−γI,dGdt=βθ+(Xx)θ−(I)−λG 1 The key parameters include the maximal expression rates α, β, and μ. These parameters encapsulate the combined strength of a promoter and its RBS, allowing each to be independently tuned. The basal degradation–dilution rate γ governs the turnover of untagged proteins, while λ denotes the higher degradation rate of tagged proteins. Biologically plausible parameter values for the nondimensionalized model, along with a detailed discussion of their interpretability, are provided in the Supporting Information (Subsection S1.1).

Functions θ±(·) are normalized transfer functions of the promoters, defined as standard Hill form with half activation constant K 1/2 = 1, namely: θ−(I)≔11+I2,⁣θ+(Xx)≔(Xx)21+(Xx)2 2

To assess the circuit’s capacity to habituate to a periodic pulsating signal, simulations are conducted starting from a steady state with no external input. A periodic external input x is introduced as an instantaneous pulse lasting for a duration Δτon, followed by an input-free interval of duration Δτoff, allowing the system to relax. The total period of each cycle is therefore Δτ = Δτon + Δτoff. The trajectory is partitioned into a collection of n consecutive intervals T = {[ti,ti+1)}i=1n, where each t i corresponds to the onset of the i-th input pulse, and each interval has a constant duration Δτ = t i+1 – t i .

An example of the time-series resulting from this minimal circuit is shown in Figure d, where the time-series for variables x, I, and G are displayed. The concentration of repressor I shows a growing trend over time, decreasing slightly during relaxation periods as its production halts while continuously decaying with rate γ. A habituation pattern is displayed by G with regular peaks displaying a decrease in amplitude with each successive input pulse.

The conditions required to guarantee the growth of the memory component can be analytically derived, given some approximations. Since a memory component is also relevant for other motifs discussed, a basic derivation is warranted (see also Supporting Information Section S2).

To this end, trajectories are partitioned into periodic intervals, each consisting of an active phase of duration Δτon, during which an external input is applied, followed by a relaxation phase of duration Δτoff, where no input is present.

Assuming that the memory component does not approach saturationspecifically, that I(t) ≪ α/γ holds throughout the entire trajectoryand further assuming that the duration of each pulse is much shorter than the relaxation period (Δτon ≪ Δτoff), the dynamics of memory accumulation can be effectively approximated as a sequence of instantaneous increments occurring at the onset of each pulse, followed by continuous exponential decay between events.

Under the impulsive approximation, the temporal evolution of the memory component is governed by the following differential equation: dIdt=αΔτon∑k=1Nδ(t−kΔτ)−γI 3 where δ(·) denotes the Dirac delta function, representing a sequence of N instantaneous inputs applied at discrete times t = kΔτ.

Assuming negligible initial conditions, memory dynamics are given by I(t)=α∑k=1NH(t−tk)e−γ(t−tk) 4 where H(·) denotes the Heaviside step function, a piecewise function defined as 0 for negative arguments and 1 otherwise, and t k = k Δτ are the times where input pulses are applied.

The change in memory over a single interval can be expressed as ΔI≈(I0+αΔτon)e−γΔτoff−I0 5 where I 0 represents the initial memory value just before the input is applied.

Since the net change in memory per cycle, ΔI, decreases monotonically as the relaxation duration Δτoff increases, and since ΔI > 0 when Δτoff = 0 (memory accumulates in the absence of extended downtime), it follows that there exists a critical threshold for Δτoff beyond which no net memory gain occurs over successive stimulation cycles. Therefore, memory accumulation is possible only if the relaxation period satisfies the inequality: Δτoff≤γ−1log(1+αΔτonI0) 6

As the number of events increases, N → ∞, the periodic steady state solution approximates: I(t)≈αΔτone−γτ1−e−γΔτ,⁣τ∈[0,Δτ) 7 oscillating with period Δτ ≈ Δτoff.

In Figure c a two-dimensional parameter space (α, γ), showing the adaptation strength, quantified as a fold change, FC, defined as the log2-transformed ratio between the peak response following the final stimulus and the peak response following the first stimulus: FC≔log2maxG(τn)maxG(τ1) 8 where τ1 ∈ [t 1, t 2) denotes the time interval from the arrival of the first stimulus up to the arrival of the second stimulus, and τ n ∈ [t n , ∞) denotes the time interval beginning with the arrival of the final stimulus and extending thereafter.

Two domains are clearly defined, with a boundary separating a phase where habituation occurs from another where habituation is not possible.

Robust habituation depends on memory I being able to hold its state between stimuli, which can be achieved by ensuring it degrades slowly (low γ). At the same time, increasing the memory production rate α makes habituation stronger by causing the system’s final response to be a smaller fraction of its initial output. However, a higher production rate also reduces the system’s peak response, creating a key trade-off: a stronger habituation strength comes at the cost of a lower maximum output. A more in-depth analysis of this trade-off is provided in Supporting Information Subsection S3.1.