Exploring the computing power of microbes that shapes the environment

Abstract

Microbes process input information into output responses through diverse genetic and metabolic mechanisms, effectively making them physical systems that compute. These computations profoundly shape the environment, from driving key chemical cycles in the soil to influencing the planet’s atmosphere. Yet the complexity of natural microbial computations remains poorly understood, including the symbolic representation of information and the underlying algorithmic principles. Synthetic biology provides tools to implement simple but effective genetic circuits in living cells, enabling human-defined computations. These are typically Boolean gates and circuits for combinatorial input processing, but they also include sequential logic, memory-based systems, analog circuits, and distributed computations in cellular consortia. Twenty-five years after the first synthetic genetic circuits were built, the field is now exploring new approaches to move closer to the computing power of natural microbes. With a focus on bacteria, this review examines both natural and synthetic functions with the aim of bridging the complexity gap between them and argues that understanding and formalizing the ways in which microbes compute may be essential for improving synthetic genetic circuitry.

Introduction

Microorganisms inhabit every environment on Earth, from soil ecosystems driven by microbial interactions and nutrient cycling, to extreme habitats characterized by temperature fluctuations, heavy metals, or unusual acidity, as well as phototrophic and aquatic systems shaped by pollutants, light, and nutrient gradients. The images in Figure 1a illustrate broad ecological contexts in which microbes integrate multiple, overlapping signals. Life constantly senses, integrates, and responds to external stimuli. Microbes rarely live in isolation; instead, they exist in complex communities, engaging in cooperative, competitive, and syntrophic interactions that govern their behavior and function. These ecosystems are influenced by physical, chemical, and biological variables, which act not as isolated triggers but as parts of an intricate information network regulating interspecies communication. Cells are not passive entities in these environments; they operate as sophisticated information-processing units, capable of integrating diverse environmental inputs into tightly coordinated physiological responses, both individually and collectively (Figure 1b). This natural ability to sense, process, and respond to environmental signals resonates with the concept of computation: living microbes compute. Over the past 25 years, this insight has inspired the use of synthetic biology as a toolkit to engineer biocomputing systems [1].

Figure 1. The complexity gap between synthetic genetic circuits and natural microbial processes.

(a) Four representative ecological contexts, chosen to exemplify the diversity of microbial habitats (from left to right, mesophilic, temperature-extremophilic, composition-extremophilic, and polluted environments). (b) Conceptualizing environments. Each niche can be represented as three interconnected layers. First, an input information layer, consisting of physico-chemical and biological stimuli that interact with one another (e.g. water limitation and salinity) and feed into the next layer. Second, a microbial network layer, where microbes sense and respond to these stimuli through highly interconnected regulatory and metabolic processes. Third, an output functional layer, encompassing the ecosystem functions required to maintain environmental balance. We propose that any environmental niche (as in panel A) can be conceptualized within this three-layered information structure of inputs, processing, and outputs. (c) Synthetic biology builds circuits from the bottom up. Each element within the microbial network (bottom) represents a complex information system, with dense protein interaction networks (middle). From these networks, synthetic biology extracts basic regulatory components — such as LacI and its cognate promoter — as the building blocks of engineered genetic circuits, exemplified here by the toggle switch (top). (d) Example of a relatively straightforward natural process: the transformation of PET (input) into energy (output) by a single bacterium. Even this simple process relies on metabolic and regulatory networks that are orders of magnitude more complex than current synthetic circuits. (e) Example of a more complex transformation: the conversion of milk (input) into kefir (output) by a consortium of bacteria and yeasts. This microbial collaboration — just one illustration of global-scale microbial processes — also far exceeds the computational capacity of existing genetic circuits.

Genetic circuits lie at the heart of biocomputation in microbial cells such as bacteria and yeasts. These circuits are DNA-based constructs, designed according to synthetic biology principles, that execute specific functions [2]. A common example is the engineering of Boolean logic operations, where genetic devices detect combinations of inputs and trigger the expression of an output protein under predefined conditions. For instance, a NOR gate is a genetic construct that senses two inputs and activates a target gene only when both inputs are absent [3,4]. Beyond logic gates, researchers have developed memory switches [5], neuromorphic circuits [6], probabilistic computing designs [7], analog processors [8], and multicellular computing through distributed circuits [9]. While engineered circuits have demonstrated remarkable successes at both fundamental and applied levels, they remain far from matching the flexibility, robustness, and computing power of natural microbial systems, which operate within open-ended and evolving environments.

This leads to what we refer to as the complexity gap between natural microbial networks and bottom-up synthetic biology. For example, a synthetic toggle switch is a relatively simple circuit that relies on regulators that themselves interact with the wider cellular machinery as part of the cell’s broader protein network; in turn, individual cells operate within microbial communities that impose additional constraints on their behavior (Figure 1c). The gap between what can currently be engineered and the layers of complexity present in nature remains substantial. In natural ecosystems, environmental inputs form a dense web of signals, which microbes integrate through regulatory and metabolic networks to generate ecological functions that stabilize the system. These layers are interconnected, with feedback loops producing emergent behaviors. An example of such behavior is the λ-phage genetic switch, which relies on layered feedback and coupling to the host to produce robust, bistable decisions. In contrast, synthetic switches capture only a simplified subset of these mechanistic features.

The engineering of genetic circuits in synthetic biology, by contrast, proceeds in the opposite direction. Beginning with highly reduced components of protein networks, such as LacI and its cognate promoter, circuits are assembled bottom-up into simplified input–output designs. This approach makes control more tractable, but it overlooks the multilayered adaptive logic that natural systems employ — that is, their capacity to integrate fluctuating or dynamically changing signals and to adjust their responses accordingly. The resulting gap is not merely one of scale but also of architecture: distributed, interdependent networks in nature versus linear, reductionist circuits in synthetic designs.

Most synthetic genetic circuits developed to date remain relatively simplistic. They typically rely on a narrow set of externally supplied chemical inducers, such as Isopropyl-beta-D-thiogalactopyranoside (IPTG) or aTc, which are stable analogs of naturally occurring molecules (IPTG: allolactose; aTc: tetracycline). These compounds were designed to improve stability, resist cellular degradation, reduce metabolic interference or crosstalk, and enable precise dose control under pristine laboratory conditions [10]. Although effective in controlled settings, and despite the growing number of studies designing and testing increasingly complex genetic circuits, such inputs fail to capture the true complexity of natural environments and therefore limit the real-world applicability of engineered systems. Moreover, synthetic circuits are most often implemented in model microorganisms such as Escherichia coli, either individually or in small consortia. Other chassis used by the synthetic biology community include the bacteria Pseudomonas putida and Bacillus subtilis, as well as the yeast Saccharomyces cerevisiae, but this limited set of species still narrows the functional diversity when compared with natural microbial communities. Therefore, large-scale environmental applications, such as bioremediation, will require a deeper understanding of biological complexity and of how microbes process environmental information, so that this knowledge can be incorporated into circuit design. Importantly, this does not necessarily imply engineering bigger or more complicated circuits, but rather designing circuits that are more efficient and effective within their domain of application.

To address these challenges, researchers are increasingly exploring how to harness naturally occurring environmental signals as inputs for genetic circuits and microbial interactions. In this paper, we explore the natural sensing and information-processing functions of microorganisms and how these capabilities can be repurposed into genetic circuits. Embracing biological complexity, such as context, noise, and evolution as design opportunities, will expand future biocomputation tools.

Beyond IPTG: expanding the input space of synthetic biology

Across the three layers of input, processing, and output into which ecosystems can be conceptualized (Figure 1b), with microbes at the central position, the complexity gap between natural and synthetic networks (Figure 1c) becomes evident. The first layer — the input network — poses a fundamental challenge for synthetic biology: expanding the repertoire of signals that can encode information. Whereas environmental niches expose microbes to a vast and diverse set of chemical and physical cues, synthetic circuits typically rely on a narrow set of standardized chemical inputs (e.g. IPTG, aTc, 3MBz). This restriction not only narrows the scope of applications but also constrains the complexity of computations that engineered systems can achieve.

Here, we highlight recent advances in integrating sensing machinery for both chemical and physical inputs into circuit engineering (Table 1), and we discuss why the ability to combine multiple inputs and capture community-level dynamics is an essential strategy for the next generation of synthetic circuits.

Table 1. Natural inputs and regulatory mechanisms adapted for synthetic genetic circuits.

Input	Natural regulation	Synthetic adaptation
Chemical signal
Carbon sources (glucose, xylose, aromatics)	Hierarchical carbon use via regulations: Crc/CrcY-Z sRNAs, ClrB, XlnR, CreA, AmyR.	Post-transcriptional control in E. coli using Crc-Hfq system from Pseudomonas [39]
Xenobiotics (i.e. plastic)	Enzyme-mediated degradation (PETase, MHETase) enables metabolic assimilation.	Engineering PET metabolism in Pseudomonas putida [15] or into bacterial consortia for plastic upcycle [16]
Heavy metals (Hg2+, As3+, Cd2+, Pb2+)	Metal-responsive repressors (MerR, ArsR, CadC) regulate transcription based on metal concentration.	Amplifier circuits for scalable metal outputs [40], biosensors [13].
Quorum sensing	Quorum-sensing molecule (AHL, AI-2, AIP) produced by its synthase binds the receptor transcription factor, controlling population-dependent gene expression.	Population control [18]
Physical signal
Light of a specific wavelength	Photoreceptors (Cph8, CcaS, UirS, proteoRhodopsin) suffer conformational changes (cofactor break or shift) → transcription, channel opening, electron transport.	Used as an input for the triggering of the production of a protein [23,24]
Ionizing radiation (X-rays, α-particles)	DNA-damage-promoter (dr0423) activating repair pathways	Radiation biosensor using β-galactosidase reporter for amperometric detection [26]
Electron flux changes	Redox-sensitive TFs (SoxRS) activate detox networks.	Electronic control of induction using pyocyanine, pSoxS, and ferricyanide redox cycling [41]
Mechanical stimuli	Second messengers (c-di-GMP, cAMP, or sRNAs) switch between motile/attachment and biofilm states.	No direct surface sensors yet; c-di-GMP sensors developed; biofilm manipulation widely engineered [34]

TF, transcription factor; AIP, autoinducing peptide.

Regarding chemical inputs sensed by microbial networks in natural environments, we classify four major groups: nutrients, xenobiotics, metals, and quorum-sensing (QS) signals.

Nutrient availability, particularly carbon sources, is a primary determinant of microbial adaptation and survival. The carbon catabolite repression (CCR) mechanism enables hierarchical tuning of gene expression depending on environmental composition, prioritizing the most favorable substrates by regulating transcription and metabolic flux [11]. Synthetic biologists have exploited CCR for optimizing sugar uptake, modulating central metabolism, and designing context-dependent regulatory circuits [12]. Also, metal sensing is widespread among bacteria, regulating essential metals involved in processes such as nitrogenase activity in nitrogen-fixing species, stress responses, and the use of metals as electron donors or acceptors. Synthetic biology has harnessed metal-sensing systems to engineer biosensors for environmental monitoring [13].

Xenobiotic sensing has been repurposed, for example, to link pesticide detection to metabolic degradation [14]. Another relevant class of xenobiotics includes synthetic polymers such as polyethylene terephthalate, where naturally occurring plastic-degrading bacteria have inspired efforts to transfer this capability into model organisms like Pseudomonas putida [15] or into engineered bacterial consortia [16]. Plastic degradation illustrates a relatively straightforward function, with a defined input and a single microbial agent as the processor (Figure 1d). Yet, the metabolic and regulatory mechanisms underlying this process are far more intricate than the mechanistic design of synthetic circuits, a level of processing that bottom-up synthetic biology has not yet achieved.

Among chemical signals, QS represents the fundamental system of intercellular communication, enabling bacteria to coordinate gene expression in response to population density [17]. In natural contexts, QS regulates diverse processes, including bioluminescence, virulence, biofilm formation, and interspecies competition. Engineered QS systems have been applied to population control [18], distributed multicellular logic [19], and have been adapted to different bacterial chassis [20]. However, QS modulation often produces pleiotropic effects, simultaneously influencing multiple cellular functions and thereby disturbing metabolic balance and energy homeostasis [17].

This latest paradigm — multicellular computing in microbial consortia — is the rule rather than the exception in natural ecosystems. A familiar household example of a domesticated natural consortia is kefir fermentation (Figure 1e), where diverse organisms coordinate through metabolic byproducts, cross-feeding, and feedback loops that couple metabolic state to gene regulation [21]. Despite centuries of empirical use, this process remains incompletely characterized, and so far, the most complex system successfully mimicking it in vitro is a three-strain consortium [22]. Both examples (Figure 1d,e), although mechanistically unresolved, are sufficiently well characterized at a black-box level to offer meaningful parallels to bottom-up synthetic biology efforts.

With respect to physical inputs — less explored in synthetic biology than the chemical ones — we highlight four key signals: light, radiation, electricity, and mechanical stimuli.

Light sensing through optogenetic systems provides clean, reversible, and tunable control of gene expression. Since the development of a red light-responsive two-component system in E. coli [23], complexity has advanced to full red, di-GMP), a secondary messenger activated by surface green, blue input systems capable of controlling three pigment outputs, effectively emulating printing with bacterial cultures [24]. Regarding the second group of inputs, identifying the machinery that directly senses ionizing radiation has proven challenging. Radiation-tolerant bacteria instead rely on global regulatory responses — such as antioxidant defenses and DNA repair pathways — to mitigate damage [25]. Nevertheless, radiation effects have been detected by coupling stress promoters to bioluminescent outputs, and more recently, the discovery of a radiation-responsive promoter enabled a biosensor platform for X-rays and alpha particles in Deinococcus radiodurans [26].

Electrical inputs are essential for maintaining homeostasis across bacterial membranes, sensed through redox-mediated regulators and ion channels that influence gene expression [27]. In synthetic biology, this capability opens opportunities for engineering electroresponsive circuits — allowing remote or real-time control of gene expression via voltage changes [28,29] — as well as for designing environmental biosensors with rapid response times [30].

Finally, mechanical stimuli — such as pressure, shear forces, or surface interactions — regulate key microbial processes, often linked to lifestyle transitions (e.g. planktonic to sessile growth), biofilm and colony [31] formation, group identity [32], and virulence [33]. Although the biology remains complex, advances in synthetic mechanosensors could allow engineered cells to function in environments where chemical or optical inputs are less accessible. For instance, cyclic di-GMP (c-di-GMP), a secondary messenger activated by surface cues, has been used to develop biosensors for its own intracellular monitoring [34], and later this type of sensor has been proposed as a tool, for example, to study antimicrobials that target this pathway [35]. Importantly, surface sensing is rarely driven by a single signal: it typically integrates multiple cues, such as physical contact and nutrient gradients, effectively functioning as a natural logic AND gate.

Choosing between chemical and physical inputs should be guided by the functional requirements of the system, such as the desired resolution of control, the degree of spatial or temporal precision, and the environmental conditions in which the circuit will function. Just as natural networks adapt their input preferences to ecological and physical constraints, synthetic designs can benefit from strategically matching input types to applications, thereby creating more robust, context-aware, and adaptable systems. This could be achieved by treating the entire host cell as the programmable substrate, moving away from circuits whose behavior depends solely on a specific input at a specific level. Instead, circuits should draw on the host context — its metabolism, resources, and mechanistic and functional features — to integrate the input signal into a richer computational function that is more resilient and robust. Context-dependent strategies of this kind are already emerging, both in the engineering of intracellular processes and in multicellular systems [19,36–38].

The ability of microbes to sense both chemical and physical cues highlights the flexibility of their signal-processing networks. Yet in natural settings, these signals rarely occur in isolation. Instead, microbes encounter overlapping gradients of input signals within communities, meaning that these signals are not homogeneously distributed in space. Communities cope with such heterogeneity through various mechanisms; for instance, cooperation, competition, and intercellular communication can facilitate information processing and shape collective responses — capabilities that may not be achievable with a single strain. Synthetic biology still faces major challenges in translating the features of natural ecosystems into engineered systems. Much of this difficulty stems from conceptualizing inputs in synthetic circuits as strong, singular signals. Shifting the focus toward heterogeneous information at the input layer would enable the engineering of multicellular systems from a more natural perspective, harnessing noise, gradients, adaptation, variability, and uncertainty.

This contrast raises a key question: if natural microbes thrive by processing noisy, dynamic, and interdependent signals, how far can synthetic biology progress while remaining dependent on the clean, uniform inputs of the laboratory — inputs that natural microbes rarely, if ever, encounter? This mismatch between the multi-dimensional inputs found in nature and the simplified inputs used in engineered circuits is a core aspect of the complexity gap, which limits circuit performance outside laboratory conditions. The issue is twofold. On the one hand, expanding the catalog of inputs — and their cognate sensing modules — would naturally broaden the design space. On the other hand, gaining a deeper understanding of how living cells process information that is, for example, noisy or uncertain, would enable the implementation of computations that extend well beyond current combinatorial logic, toward more sophisticated models of computation.

Beyond Boolean: synthetic circuits in space, time, and context

The engineering of genetic circuits was initially driven by principles of standardization and modularity to achieve maximum predictability [42]. While these concepts remain central to the field, there is now a growing tendency to emulate natural systems in their ability to handle complex information encoding, including context dependence, metabolic burden, gene expression noise, evolution, and spatiotemporal asymmetries. In other words, synthetic circuitry increasingly seeks to build on what natural systems do best. These features represent key dimensions that must be addressed to bridge the complexity gap between synthetic and natural computations. While there are examples in each area, as outlined below, significant challenges remain to make substantial progress. Of course, gaps in our knowledge — or, more accurately, the lack thereof — are a major barrier. Equally important is drawing on insights from the theory of computation, which already provides guidance on how to move beyond simple Boolean logic.

Rather than operating through absolute threshold-based logic and idealized on/off states, cells naturally integrate multiple graded inputs that are distributed asymmetrically in space and time. Moreover, they generate graded, adaptive responses. Synthetic circuits that exploit spatial asymmetries to process non-uniform signal distributions, differential cellular states, and diffusion include (Figure 2a): radial gradients induced by arabinose [43]; Turing and Hopf patterns producing rings [44] or spots [45]; and checkerboard arrangements generated by optogenetic lateral inhibition [46]. Temporal control is equally intrinsic to biology. For example, the ftsZ gene encodes a cell cycle-dependent protein responsible for septum formation during division, whose oscillating concentration is governed by several promoters in a complex regulatory network (Figure 2b). Time also plays a central role in synthetic systems: carbon sources decay as they are consumed, while quorum-sensing effects require the gradual accumulation of sufficient N-Acyl homoserine lactone signals. Such dynamics can disrupt Boolean circuits, which depend on idealized binary states. We argue, however, that spatiotemporal control is fundamental for advancing the field; the engineering of patterns, optogenetics, or pulse-based and probabilistic [47] computing approaches are promising endeavors toward that goal.

Figure 2. Synthetic circuits in challenging scenarios.

(a) Circuits addressing spatial asymmetries and heterogeneity: a radial gradient [43], Turing and Hopf patterns [44, 45], and checkerboard arrangements [46]. (b) Inputs with temporal dynamics. Examples include cell cycle-dependent oscillations of ftsZ expression, carbon sources acting as decaying inputs as they are consumed, and QS signals that depend on the gradual accumulation of AHL molecules. (c) Distributed computations across strains. Kusumawardhani et al. developed an M13 phage-based communication channel to regulate gene expression remotely [9]. Optogenetic control of antibiotic resistance has been used to stabilize co-cultures [49]. Crespo-Roche et al. engineered an interkingdom consortium in which fungal metabolism provides intermediates for P. putida bioplastic production [50]. Kong et al. demonstrated pathway reconfiguration in consortia to enable programmable social interactions [51]. (d) Contextual dependencies in circuit function. Alejaldre et al. [56] showed how toggle-switch activity changes with chromosomal integration site. Tin Chat Chan et al. [74] concluded that combining RBS tuning with host background enables major improvements in toggle performance. Tellechea-Luzardo et al. [57] highlighted environmental context by testing a naringenin sensor in different media, revealing strong performance shifts. (e) Strategies to mitigate metabolic and genetic burden. Guan et al. added a secondary operon for TF binding, reducing stress [75]; Ceroni et al. engineered feedback to improve growth and host capacity [61]; and Williams et al. diversified populations so only a subfraction expressed the circuit [62]. (f) Circuits that avoid or exploit noise. Negative feedback loops suppress noise, while toggle switches can be engineered for reduced variability. Conversely, some designs leverage stochasticity, such as Elowitz’s repressilator [66, 67]. (g) Circuits incorporating evolutionary dynamics. Strategies include securing gene function through co-regulation (bidirectional promoters, shared elements), promoting beneficial mutations while avoiding loss-of-function (overlapping genes), and buffering instability via redundancy [73].TF, transcription factor

Multicellular computations [48] within consortia further exploit collective behaviors in spatiotemporal contexts (Figure 2c). For example, phage-mediated communication has been used to transfer sgRNAs between strains, enabling programmable distributed circuits [9], while optogenetic feedback has been applied to dynamically regulate co-culture composition by controlling antibiotic resistance in light-responsive strains [49]. Expanding beyond bacteria, synthetic interkingdom consortia show how fungal quorum-sensing and metabolism can support P. putida in producing bioplastics from agro-residues [50]. Likewise, bacteriocin-mediated pathway reconfiguration illustrates how signaling and antimicrobial activity can be combined to engineer defined social interactions between strains [51]. These and other strategies, such as engineering cross-feeding to stabilize consortia [52], rarely extend beyond two strains — a clear limitation when compared to the diversity and complexity of natural microbiomes.

Natural systems can involve highly diverse communities, with numerous species interacting through complex signaling. Recent advances have begun to scale synthetic efforts in this direction. For instance, a six-strain consortium was engineered to solve a chemically encoded 2 × 2 maze problem [53]. An even more ambitious study implemented a 2-bit encoding system across 66 bacterial strains, each interacting — albeit in a controlled setup — and contributing to a global computation. This platform coordinated more than 110 logic gates, transmitting intermediate states across strains through small-molecule signaling [54].

Even at the single-cell level, the reductionist view that circuit function is determined mainly (or even solely) by DNA sequence has proven inadequate. In practice, performance emerges from the interplay between the genetic sequence, the host context, and the surrounding environment [55]. Circuit design increasingly embraces this complexity. Recent efforts to expand the library of genetic logic gates to account for contextual dependencies (Figure 2d) — including plasmid vectors, host chassis [37], or chromosomal integration sites [56] — show that these dependencies can already be leveraged to select for desired functional outputs. Environmental factors further modulate circuit performance metrics [57]. Context is thus a highly complex but essential element that living systems exploit, and characterizing host organisms is key for building circuits from the bottom up. While E. coli remains the most studied chassis, its well-characterized biology does not guarantee suitability for every application [58]. Host-specific factors — including integration sites, intercellular interactions, and metabolic state — can generate non-orthogonal behavior and limit portability [38]. To date, most circuits have been tested in only one or a few species, and transferring them to new hosts often raises interoperability issues that require extensive fine-tuning [59]. One strategy has been to isolate circuits from their context [60], a valid and sometimes necessary approach when predefined functions are required. However, this comes at the expense of exploiting the variability and adaptability that natural systems derive from context.

A central aspect of host context involves resource allocation — how cellular resources are distributed among competing functions — and the burden imposed by synthetic circuits. Natural networks have evolved to operate near physiological resource limits to sustain essential processes, whereas in engineered systems, circuits often overload the cellular machinery, leading to metabolic burden [61]. Synthetic biology has addressed the burden through two main strategies (Figure 2e). One approach is to incorporate dedicated modules that allow circuits to autonomously balance their output before resource depletion becomes critical [36,61]. Another strategy uses integrases to control expression, confining burden to a subset of differentiated strains, while the remaining propagation strains maintain the circuit without significant fitness costs [62].

Living organisms rarely compute with narrow, deterministic responses. Instead, gene expression is often noisy, shaped by multiple factors and extending from transcriptional bursts to cytoplasmic composition, with profound consequences for behavior [63]. This molecular noise underpins bet-hedging strategies: transcription occurs in stochastic bursts, generating intrinsic noise, which can be amplified or buffered by translation dynamics [64] and protein stability. Noise propagates through regulatory cascades, producing correlated fluctuations — extrinsic noise — that enables coordination among genes [65]. Synthetic circuits have both exploited and mitigated these properties. For example, the repressilator demonstrated delay-driven oscillations with noisy molecular signaling [66], while noise has also been shown to broadly modulate circuit dynamics [67]. Negative feedback loops have been used to suppress variability [68] (Figure 2f) or to reduce expression at the off state to (almost) absolute zero [69]. Rather than being treated as a problem, intrinsic and extrinsic noise should be considered fundamental design elements for bio-computation [70].

Closely linked to noise, evolution is another major source of unpredictability in engineered circuits. Yet, it can also be harnessed, as in assisted laboratory evolution, widely used to optimize traits such as sucrose metabolism [71], ethanol tolerance, and growth rates [72]. The challenge is that once optimal performance is reached, further evolution often destabilizes circuits. To counter this, strategies reduce selective pressures against engineered parts, for instance through bidirectional promoters, overlapping genes, shared genetic elements, or redundancy [73] (Figure 2g). While it is generally assumed that evolution cannot be stopped — only slowed — this raises a provocative question: instead of fighting unpredictability, could evolution be domesticated as a controllable circuit design feature?

Discussion: closing the complexity gap

A central vision of synthetic biology is to treat life itself as programmable matter. Realizing this vision requires un-locking the full computational potential of living systems, thereby enabling transformative applications ranging from large-scale Earth bioremediation [76] to synthetic life designed for space exploration [77]. The field is already advancing toward increasingly complex forms of computation in key areas: engineering circuits that remain functional across diverse environments [37,57]; designing constructs that harness, rather than minimize, inherent biological characteristics such as noise [63] and evolution [73,62]; and developing multicellular computing systems with distributed and coordinated interactions [19]. Although bio-computation is moving in this direction, it remains far from robust real-world applications — particularly within ecosystems — as major challenges persist. In this review, we have highlighted that even relatively simple genetic circuits can be designed to operate under constraints such as spatial heterogeneity, host context, and molecular noise — factors that contribute to the persistent gap between the behavior of natural systems and the performance of synthetic designs, and which will continue to pose obstacles unless we actively incorporate them into our engineering approaches.

Complexity-both computational and systemic — gives cells a fundamentally different way of computing information, one that may enable more robust and versatile computations [1,7]. Embracing this potential requires acknowledging the inherent non-linearity of life: evolution, noisy gene expression, and graded signals that shape dynamic interactions. Integrating biological complexity also calls for reinterpreting biocomputation [78] — not as a centralized, deterministic process, but as a distributed, context-aware, and largely liquid phenomenon [79]. Rather than treating noise, context, or evolution as nuisances, they can serve as entry points for deeper insights into natural networks, guiding us toward designs that work with, rather than against, the grain of biology. Emerging studies already suggest that such variability can be harnessed for adaptive design [74], supported by quasi-mechanistic models capable of predicting circuit behavior across hosts [80].

Although living systems can never be fully predictable, biocomputation still depends on simplifying life into predictable parts and circuits. Resources such as libraries of well-characterized ribosome binding sites exemplify how standardization remains indispensable for practical applications [74]. At the same time, progress also requires acknowledging that some aspects of natural systems may not be fully predicted or simplified. These two approaches are not contradictory but complementary, and together they define the path toward closing the complexity gap, the distance between bottom-up genetic circuits and the full functional scope of microbial ecosystems. The challenge, then, is to balance predictability with complexity, and to learn how to design at the very edge of uncertainty — where the most transformative innovations are likely to emerge.

Data Availability

No data were used for the research described in the article.