# Ecological risks of synthetic biology Horizontal gene transfer, fitness cost, and the difference between auxotrophic and synthetic-codon containment — with an honest reading of what a measured escape frequency does and does not mean. Why most engineered constructs disappear from a wild population by themselves, and why the containment numbers that look reassuring are laboratory numbers. Source: https://en.bioecon.ru/docs/services-governance-capital/regulatory-legal/ecological-risks-of-synthetic-biology/ Updated: 2026-09-06 The ecological question about an engineered organism is not whether it can survive outside containment but whether its construct persists and spreads. Those are different questions with different answers, and conflating them produces both false alarm and false reassurance. ## Why most constructs are lost An engineered construct is a metabolic burden. Transcribing and translating heterologous genes consumes ribosomes, precursors and energy that are not then available for growth. In the laboratory this cost is hidden because an antibiotic or a defined medium selects for the construct; in a wild population there is no such selection, so a cell that loses the insert — by plasmid segregation, by recombination, by a frameshift in the cassette — outgrows one that keeps it. The result is dilution: the construct falls in frequency at a rate set by its fitness cost, and burdened engineered strains are typically outcompeted by wild relatives. This is a rate argument, not a guarantee, and there are two ways it fails. Compensatory evolution can reduce the cost of carrying a plasmid over a few hundred generations without the plasmid being lost. And a construct that confers any advantage in some environment — a resistance determinant, a novel catabolic pathway — is under positive selection there, in which case dilution reverses. ## Horizontal transfer moves the gene, not the organism Bacteria exchange DNA by conjugation, natural transformation and transduction, across species boundaries and at rates that depend strongly on the vehicle. A mobilisable plasmid with an origin of transfer is the high-risk carrier; a chromosomally integrated cassette without flanking mobile elements is far less so. This is why containment design pays attention to what a gene sits on: killing the host does not retire a gene that has already moved. ## Two kinds of containment, and what separates them Auxotrophic containment deletes an essential biosynthetic gene so the strain dies without a supplement — thymidine, or diaminopimelate for peptidoglycan cross-linking. Its weakness is that the missing function is available in the environment. The strain can be cross-fed by neighbouring organisms or lysed biomass, and the deletion can be complemented by horizontally acquired DNA, because the required molecule is one that ordinary biology already makes. Synthetic containment changes the requirement to something biology does not make. In genomically recoded *Escherichia coli*, the UAG codon is freed of its stop function and reassigned to a non-standard amino acid, which is then written into essential proteins; without the synthetic amino acid, those proteins truncate. Rovner and colleagues and Mandell and colleagues reported such strains in *Nature* in 2015 with escape frequencies below the detection limit of their assays, in the region of 10⁻¹² per cell. Nothing in the environment supplies biphenylalanine, and no horizontal transfer confers it. ## Reading an escape frequency honestly A figure like 10⁻¹¹ is an assay result: a known number of cells plated without the supplement, incubated for a defined time, escapers counted. It bounds one failure mode under one condition. A release involves cell numbers many orders larger, over seasons rather than days, in the presence of donors of DNA the assay never contained, and with selection pressures the laboratory did not apply. The number is a real measurement and a lower bound on safety, not a probability of ecological failure. The same reservation applies to gene drives, where the dominant known failure is internal: repair by non-homologous end joining generates drive-resistant alleles at the target site, so drive efficiency depends on choosing a target sequence so conserved that resistant variants are themselves unfit.