The Moore's Law of Synthetic Gene Circuits
Why biology's circuits stopped scaling and what might change that
Introduction
In the late 1950s, on either side of a corridor of the Institut Pasteur, seemingly unrelated efforts on E. coli’s sugar metabolism and the propensity of a virus to reproduce would converge into the discovery of the first two “gene circuits”.
Bacteriophages are viruses that infect bacteria and inject their genetic material, hijacking the cell’s machinery to produce new viruses that burst out of the cell. André Lwoff dedicated his career to understanding the regulation of this process that, at the time, was not even known to be genetic as DNA would gain prominence in biology only from the discovery of its structure and the central dogma, 1954 and 1957. Jacques Monod instead had written his doctoral thesis on the growth of E. coli in different culture conditions, first characterising diauxic growth (growth on two carbon sources). While Monod joined Lwoff’s lab in 1945 to continue his research on the switch in the growth rate of E. coli when exposed to different sugars, François Jacob joined as an aspiring doctoral student in 1950 to pursue Lwoff’s main interest of elucidating the trigger that would cause bacteriophages to replicate and kill their host.
They soon realised that they were answering a similar question: what regulates the turning on and off of events in cells? Their joint efforts led to a publication in 1961 titled “Genetic Regulatory Mechanisms in the Synthesis of Proteins”, where they outlined how a set of genes could be activated or repressed together by a “master switch” or inducer. Similarly to an electrical circuit, where a switch can allow or stop the flow of electrons towards a lightbulb, an inducer could allow or repress the flow of genetic information or gene expression. Jacob and Monod had uncovered the first gene circuits. The duo’s discovery was not only worth a Nobel prize in 1965, as it kickstarted a revolution in our understanding of cellular events. In scientists’ minds, these were now an almost mechanical interplay of proteins interfering and shaping gene expression to regulate protein synthesis and, by extension, the cells’ phenotype.
Since the days of Jacob and Monod, scientists have discovered ever larger gene circuits regulating all aspects of the cell, such as the cyclic AMP receptor protein that interacts with more than ~180 genes to regulate catabolism in E. coli and the Myc gene regulatory network, which plays a vital role in the cell cycle and comprises ~3000 genes. Meanwhile, synthetic biologists started building synthetic versions of gene circuits to engineer specified cellular functions, first with two genes and growing over the first twenty years since the field’s inception.
The excitement for synthetic biology triggered a rush for new parts, connections and functionalities. A similar type of excitement in the 1950s for digital computing, circuits of transistors instead of genes, was followed by Moore’s law, the scaling of transistors on chips with a doubling time of 2 years. After 25 years of synthetic biology, the largest synthetic gene circuit in a single strain was 13 genes, and 63 when distributed across strains functioning as a digital display. If we exclude the “display” circuit, split between seven strains, and consider the largest intracellular gene circuit, the doubling time is ~7.2 years, far from a Moore’s law of synthetic gene circuits.
Synthetic gene circuits enable us to encode complex logic into living organisms that we can use to reroute their metabolic pathways, control their development, or enhance them with new functionalities. They have already been applied to CAR-T therapies to control receptor expression, bacteria for complex environmental sensing, and fungi for the bioproduction of molecules that organic chemists have long struggled with. However, synthetic circuits are still trailing behind their natural counterparts in size and complexity.
The tallest hurdle for synthetic gene circuits has been the interconnected nature of cells and organisms, where changes can have unintended ripple effects, disrupting the circuit or its host. To scale beyond natural limits imposed by the availability of resources and evolutionary tendencies, we need to shift away from the predictable incremental addition of carefully modelled parts and motifs to existing, limited systems. Instead, we should embrace bottom-up cell engineering, from membranes and organelles to genomes, leading to systems carefully designed with their specific function in mind. Foundational genomic models could act as a catalyst by enabling us to construct parts with built-in resource considerations while streamlining genetic design, accelerating our paths to genome-sized synthetic gene circuits.
History
Over the three decades following Jacob and Monod’s discovery, exceptional progress was made in understanding genetic regulation at the molecular level. The introduction of recombinant DNA in 1972, cut and paste for DNA, DNA sequencing in 1977, reading DNA, and PCR (polymerase chain reaction) in 1985, photocopying DNA, are among the breakthroughs that enabled us to characterise these natural systems in ever finer detail. Armed with these techniques, scientists exponentially grew our knowledge of natural gene circuits over the following two decades. At the close of the millennium, this crescendo of biological information culminated in an idea at the intersection of molecular biology, systems biology and control engineering. The proponents were a diverse group of physicists, engineers, biologists and computer scientists who believed that, by building on previously characterised biology, we could engineer new biological functions from scratch in the lab.
Among them was Jim Collins at Boston University. After struggling to find someone willing to “play engineer” with biology, he met Tim Gardner, a mechanical engineer from Princeton, who took on his project to build a switch with transcriptional repressors. A couple hundred miles south, Michael Elowitz, then a physics graduate student at Princeton joined forces with Stanislas Leibler, eminent systems biologist, to work on a project laughed at by prominent molecular biologists which utilised transcriptional repressors to generate oscillations. In 2000, the two teams published two-part and three-part synthetic gene circuits used to control the expression of a fluorescent protein in a mathematically predictable manner. The toggle switch published by Jim Collins consisted of two inputs, two regulatory parts and one output, while the oscillator had one input, three regulatory parts and one output. Interestingly, both synthetic gene circuits used the lacI and the CI repressors, the “switches” from the first natural gene circuits studied by Jacob and Monod, but repurposed for a novel function.
As with every great discovery, it took a couple of years for the field to understand the watershed caused by these seemingly futile biological flickers. Then, a flurry of papers emerged, exploring the possibilities offered by the concept of gene circuits and what simple mathematical models suggested to be possible. The first decade of gene circuits saw standardised DNA assembly methods that increased iteration speed, predictable pattern formation in E. coli cell populations, entire metabolic pathways reconstructed in yeast, and some examples of gene circuits in complex eukaryotes such as mammalian cells. In the second decade, while circuits kept growing, aided by CAD (computer-aided design) software for E. coli circuits and the newly discovered CRISPR-Cas system, they also started to be applied in the real world with a fully biosynthetic artemisinin pathway (a malarial drug), the first logic-gated CAR-Ts (cell-based immunotherapies), and carbon fixation in E. coli for environmental remediation.
Gene circuit complexity peaked in 2020 when Christopher Voigt, a pillar of the field, published a 63-regulator circuit distributed between seven different bacterial strains. The coordinated action of these seven modified strains replicated a digital display with four molecules as input, and a visual representation of their binary-encoded message as output. Of relevance, the gene circuit’s design was fully automated, thanks to the CELLO software suite previously developed by the same lab, and the authors stressed how manually designing and testing combinations would have been nearly impossible. The paper is a hallmark of the shift in the field from computers being used to build small mechanistic models to aid experiments to fully integrated computer-aided biological design.
Despite these advances in automated gene circuit design, as of 2026, more than halfway through the third decade of synthetic biology, the focus has gone from growth in size towards the reliable application of existing synthetic gene circuits. In the medical field, advanced logic gates and distributed computing techniques were applied to CAR-T cell therapies, improving safety and efficacy. An electrical protein switch, fused to an electrochemical sensor, was deployed in rivers to sense environmental thiosulfate pollution. In bioproduction, an engineered yeast with 56 genetic edits led to the biological production of two fundamental chemotherapy precursors globally in short supply. Nonetheless, the growth rate of synthetic gene circuit size has undoubtedly slowed down in the last few years, and we are starting to understand why.
Molecular constraints
Since the 1980s, the term “metabolic burden” has been used to describe how genetically modified bacteria cope with the physiological changes caused by expressing foreign DNA. In 1990, the number of plasmids expressing foreign proteins was found to directly affect the growth of bacterial hosts. However, these issues were not significant obstacles, as most genetic engineering at the time involved overexpressing a single gene in an easily tractable host. The introduction of synthetic gene circuits has increased the importance of the number and order of foreign genes expressed simultaneously, and their interactions, to achieve the desired results. Interactions between foreign genes have become particularly significant. For example, in the functioning of a toggle switch or an oscillator, an unexpected decrease in the expression of one gene could disrupt the entire system.
In the 2010s, a few influential papers pointed to the availability of ribosomes as the resource bottleneck when expressing foreign genes in bacteria, initially with mathematical models and later with “capacity monitors” and other sensors. Capacity monitors, partly developed in my PhD lab, consist of a fluorescent protein resource-coupled to a gene of interest by expression in the same cell with the same transcriptional and translational machinery. Changes in the monitor upon the addition or removal of genes that do not directly interact with it signify a shift in the availability of resources for its expression. The same concept was applied to assess resource shortcomings in protein degradation in bacteria and gene expression in mammalian cells and to devise mitigation strategies that allocate resources in gene circuits to achieve the desired functionality. RNA-seq and high-throughput approaches such as Perturb-seq allowed us to look at how single gene alterations in wild-type cells impact the expression of the rest. Unsurprisingly, this method highlighted the interconnectedness of the cell by showing how the expression of even seemingly unrelated genes is interdependent. Due to the reliance of synthetic gene circuits on cellular infrastructure and resources, and the extremely tight regulation of endogenous gene expression and metabolism, they cannot be siloed and their design has to account for the totality of the host cell.
Even though several control mechanisms based on feedback emerged to mitigate the resource limitations of synthetic gene circuits, they are plagued by reductions in baseline expression levels and complicated circuit designs. Furthermore, these mitigation strategies are not one-size-fits-all, so they need to be redesigned when their corresponding circuit is modified or expanded. While these are constantly improving, the fact that any synthetic gene circuit larger than a threshold, which is organism and expression-level dependent, would need a resource control mechanism for each new part, turned the idea of scaling gene circuits into a pipe dream.
Present and a vision for the future of gene circuits
A world of host-reliant gene circuits
The first gene circuits were all designed for model organisms such as E. coli, Saccharomyces cerevisiae, Arabidopsis thaliana and HEK293 cells. The advantage of well-characterised hosts was that one did not need to think as much about the genes’ transcription and translation, whether they would localise to the same intracellular compartment, or if the correct metabolite or reaction intermediate was present in the cell. Similar to how we use libraries when programming, which is often not the most computationally efficient solution to our requirement, model organisms allow us to be quicker and to work at a higher level of abstraction. Lastly, designing and building synthetic gene circuits for and in well-characterised hosts lets us better understand failure modes due to the accumulated knowledge about the system.
On the other hand, introducing synthetic gene circuits in an organism that has gone through aeons of evolution to optimise its genetic program is like running a new application on a machine already at 99% capacity. Recent studies have modelled quantitatively that, in bacteria, only ~10% to ~20% of the expression budget can be allocated to foreign proteins. In mammalian cells, it was reported that depleting the cells of genes that encode for “resource-expensive” proteins augments the expression of foreign genes. The competition for expression resources, which arises from the host’s limitations, points towards a model of the cell as one gene circuit which a synthetic one can interact with. The synthetic gene circuit regulates its genes directly and the ones of the cell indirectly by sequestering resources away from them. In return, the dynamic changes in the cell’s expression patterns would indirectly affect any synthetic gene circuit introduced into the host. It has even been shown that changes in the host’s physiology caused by the expression of synthetic gene circuits can loop back and alter the dynamics of the circuit itself through a mechanism called growth feedback.
The cell can then be considered one gene circuit. If we accept this deduction together with the assumption of limited cellular resources and their rate of renewal, then the cell will have an inherent ceiling to the number of foreign functions that can be performed within it at once. Consequently, we might have limited capacity to engineer larger gene circuits within existing hosts in their wild-type form. More complex synthetic gene circuits risk altering the cell’s physiology and hampering its growth and viability. In some cases, this detrimental effect on the cell’s physiology forces the choice of splitting up the circuit, such as in the 63-part digital display circuit implemented in seven strains. However, even this solution, if scaled, leads to more issues related to the dynamic interactions of strains in co-cultures.
A future of synthetic organisms
We have witnessed through the evolution of gene circuits that their increase in size corresponded to new engineering capabilities and practical applications. Three-component circuits led to an oscillator, while six allowed for a cellular event counter. Sixty-three components were required for a replica of a digital display, while 56 targeted edits of a natural circuit produced a difficult-to-synthesise chemotherapeutic agent. Although challenges relating to resource competition and crosstalk between components have arisen, I believe further scaling synthetic gene circuits should be a north star for the field. Compounding on our ability to build synthetic biological functions enriches our knowledge of ourselves and the biology surrounding us, shifting the horizon of what is considered possible for living systems. To continue on this exponential path, we first need to clear out a road fraught with resource limitations and unknown interactions.
Synthetic genomics, minimising, refactoring, and recoding genomes, offers an avenue to free resources which could then be re-employed for synthetic functions. In 2010, the J. Craig Venter Institute synthesised the first complete bacterial genome, M. mycoides JCVI-syn1.0, achieving the most impressive feat in large-scale DNA cloning. By 2016, the same group had pared this down to JCVI-syn3.0, a minimal genome of just 473 genes. More recently, the Sc2.0 consortium completed the synthesis of all sixteen S. cerevisiae chromosomes plus a synthetic tRNA neochromosome, with over half the synthetic genome now consolidated in a single viable strain. Complementarily, the synthetic cell field studies biological membranes and how biological function can be encapsulated in or actuated by synthetic analogues. Fine-tuning the size, shape, and chemistry of cellular membranes, from organelles to the plasma membrane, allows for an artificial spatial organisation of resources and segregation of functions, which could alleviate cross-talk and interference in large synthetic gene circuits. Moreover, functions that in a cell are biological in a synthetic cell might be carried out by its modified membrane chemistry; for instance, not encoding movement, chemotaxis, adhesion, or the cytoskeleton in the genome would free up resources to engineer new functions. Some of these “tricks” have already been proven possible, including movement, partial regeneration and alternative cytoskeletons.
Advances in synthetic cell components and genomes will allow for streamlined cells with the capacity to host and maintain designer genetic programmes. The following obstacle will be to design systems encompassing regulatory proteins, RNAs, and DNA, accounting for indirect interactions, functional promiscuities and host compatibility. Fortunately, the intersection of automation, high-throughput and single-cell methods is allowing us to collect ever greater datasets on biological functions from genomics to transcriptomics and metabolomics. A recent example is the CLASSIC platform, which profiles over 100,000 gene circuit designs in a single experiment in human cells, generating composition-to-function maps that ML models can learn from to predict circuit behaviour across vast design landscapes, as well as the interactions of the different components.
The reducibility of biology to something akin to language favoured the application of large-scale machine-learning models, which exploit patterns and statistical properties of natural language, to interpret ever-growing biological datasets. In late 2024, the first version of Evo, a 7-billion-parameter ‘foundational’ DNA large language model trained on prokaryotic genomes, demonstrated the ability to generate transcriptional units encoding proteins predicted to express and fold appropriately. Within months, Evo 2 scaled to 40 billion parameters and a context window of one million tokens, trained on 9.3 trillion nucleotides spanning all three domains of life. Mechanistic interpretability studies revealed that Evo 2 had autonomously learned to recognise exon-intron boundaries, transcription factor binding sites, and protein secondary structure from raw sequence alone. These models seem to be learning fundamental biological properties, such as the structure of transcriptional units or requirements for housekeeping genes in a synthetic genome, which they were not explicitly taught. This implicit learning may be precisely what makes foundational models suited to scaling gene circuits. A telling precedent comes from protein language models: mutations favoured by ESM2 tend to be stabilising and functional, even without explicit optimisation for these properties, likely because the model has internalised the biophysical constraints that natural selection has already filtered for in ways that are difficult to rationally codify. If DNA foundational models capture equivalent rules at the genome level, what a functional operon looks like, how regulatory elements interact across distances, and what makes a circuit robust to host context, they could disrupt the trial-and-error that has bottlenecked circuit design for decades. If benchmarked on specific functions, Evo generally underperforms compared to specialist models; however, the breadth of biological ‘understanding’ it achieved, and the pace at which it is advancing, makes me wonder in what ways the next generation will surprise us. Molecular biology foundational models might be the natural successor to computer-aided biological design software in the era of the synthetic genome, the synthetic cell, and the beginning of Moore’s law for synthetic gene circuits.
A thought experiment: small symbiont to intelligent life in a decade
What would the world look like if we could achieve Moore’s law of synthetic gene circuits and double the components of a gene circuit annually? In 4 years, we could engineer something of the size of the smallest known genome, the genome of Carsonella ruddii ~180 genes, a symbiotic bacterium. A small synthetic symbiont could control insects for complex environmental engineering or act as a probiotic for microbiome engineering. In 5.2 years, we would reach the size of Mycoplasma genitalium, ~500 genes, a human pathogen and a widely adopted minimal genome model. M. genitalium is also not autonomous, but it can colonise mammalian cells; therefore, a genome of this size could serve as a cellular backpack for animals and humans, providing extra functions such as resistance to viral infections and alternative metabolic pathways. By the following year, we would already have a bottom-up, free-living, bacterium-like organism, as the smallest known has ~1000 genes, and this could serve for autonomous environmental monitoring, remediation, and as a medicinal nanobot. After 8.5 years of exponential growth, we could reach the size of Saccharomyces cerevisiae ~6000 genes, known as Baker’s yeast and “cell factory”. Baker’s yeast is the prime vector for metabolic engineering, which consists of using the cell’s metabolism to perform all kinds of organic chemistry. With an organism of that size highly optimised for specific reactions, we could “print” microreactors, enabling all sorts of chemistry in a programmable and decentralised manner. Around the 10th year, we would reach the size of the human genome, an intelligent life form capable of incredible technological developments; imagine the possibilities and what the following decade would hold.
Thanks to Niko McCarty and Kush Desai for edits and comments!
References
1. Jacob, F. & Monod, J. Genetic regulatory mechanisms in the synthesis of proteins. J. Mol. Biol. 3, 318–356 (1961). https://doi.org/10.1016/S0022-2836(61)80072-7
2. Gardner, T. S., Cantor, C. R. & Collins, J. J. Construction of a genetic toggle switch in Escherichia coli. Nature 403, 339–342 (2000). https://doi.org/10.1038/35002131
3. Elowitz, M. B. & Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature 403, 335–338 (2000). https://doi.org/10.1038/35002125
4. Paddon, C. J. et al. High-level semi-synthetic production of the potent antimalarial artemisinin. Nature 496, 528–532 (2013). https://doi.org/10.1038/nature12051
5. Nielsen, A. A. K. et al. Genetic circuit design automation. Science 352, aac7341 (2016). https://doi.org/10.1126/science.aac7341
6. Shin, J. et al. Programming Escherichia coli to function as a digital display. Mol. Syst. Biol. 16, e9401 (2020). https://doi.org/10.15252/msb.20199401
7. Zhang, J. et al. A microbial supply chain for production of the anti-cancer drug vinblastine. Nature 609, 341–347 (2022). https://doi.org/10.1038/s41586-022-05157-3
8. Ceroni, F., Algar, R., Stan, G.-B. & Ellis, T. Quantifying cellular capacity identifies gene expression designs with reduced burden. Nat. Methods 12, 415–418 (2015). https://doi.org/10.1038/nmeth.3339
9. Ceroni, F. et al. Burden-driven feedback control of gene expression. Nat. Methods 15, 387–393 (2018). https://doi.org/10.1038/nmeth.4635
10. Gibson, D. G. et al. Creation of a bacterial cell controlled by a chemically synthesized genome. Science 329, 52–56 (2010). https://doi.org/10.1126/science.1190719
11. Hutchison, C. A. III et al. Design and synthesis of a minimal bacterial genome. Science 351, aad6253 (2016). https://doi.org/10.1126/science.aad6253
12. Sc2.0 Consortium. Design, construction and characterisation of synthetic yeast chromosomes. Cell & Cell Genomics collection (2023). https://www.cell.com/consortium/synthetic-yeast-genome
13. Schindler, D., Walker, R. S. K. et al. Design, construction, and functional characterization of a tRNA neochromosome in yeast. Cell 186, 5237–5253 (2023). https://doi.org/10.1016/j.cell.2023.10.015
14. Goold, H. D., Walker, R. S. K. et al. Assembly and debugging of a synthetic Saccharomyces cerevisiae chromosome XVI. Nat. Commun. 16, 995 (2025). https://doi.org/10.1038/s41467-025-56082-6
15. Nguyen, E. et al. Sequence modeling and design from molecular to genome scale with Evo. Science 386, eado9336 (2024). https://doi.org/10.1126/science.ado9336
16. Brixi, G., Durrant, M. G. et al. Genome modeling and design across all domains of life with Evo 2. bioRxiv (2025). https://doi.org/10.1101/2025.02.18.638918
17. Gorton, L. et al. Interpreting Evo 2: Arc Institute’s next-generation genomic foundation model. Goodfire Research (2025). https://www.goodfire.ai/research/interpreting-evo-2
Nice piece. I like the printing factories idea.
i suspect these attempts to graft electrical engineering logic into biological systems that fundamentally don’t run on that OS will soon prove to be time wasted exploring the wrong trail