Building-in biosafety for synthetic biology

Horizontal gene transfer

Containment mechanisms built into GMMs, based on the Asilomar recommendations, broadly perform two tasks: (i) preventing the spread of recombinant and synthetic DNA to other organisms by horizontal gene transfer (HGT) or (ii) preventing the engineered organism from overrunning or polluting a habitat. The issue of HGT is especially relevant for microbial systems, as it is a common and somewhat uncontrolled trait throughout the microbial biosphere (Dröge et al., 1998). HGT mainly occurs by transduction (active transfer via bacteriophages), conjugation (active transfer via pili) and transformation (sequence-independent uptake of free DNA from the environment) (Davison, 1999). While it is possible to engineer solutions to prevent the active mechanisms of HGT, negating transformation is a more complicated challenge (Thomas & Nielsen, 2005). Natural cell death and lysis mean there is a continual presence of free DNA in the environment, with estimates of up to 1 µg nucleic acid per gram of soil and 80 µg per litre of marine water (Lorenz & Wackernagel, 1994; Nielsen et al., 2007). Depending on particular conditions, extracellular DNA can be detected months after being placed in the environment (Nielsen et al., 2007) and extracellular DNA is actively assimilated (for nutritional or genetic utilization) by many Gram-positive and -negative bacteria (Lorenz & Wackernagel, 1994; Thomas & Nielsen, 2005) as well as some unicellular (Hall et al., 2005) and multicellular eukaryotes (Boschetti et al., 2012). Thus, even GMMs programmed to ‘self-destruct’ pose an environmental risk, as their genes can potentially be scavenged by other cells after they have died.

Monitoring rates of HGT in the field is challenging due to the large sample sizes or long time frames required for rare transformants to reach assayable populations (Townsend et al., 2012). From studies so far the good news is that HGT events from GMMs are nearly always deleterious and that the natural transformation frequency of microbes in soil is itself less than 1×10⁻⁷ per bacterium exposed (Nielsen & Townsend, 2004). Moreover, extracellular DNA only seems to be capable of transformation for a matter of hours to days post-release into the environment despite longer retention times (Nielsen et al., 2007). Despite these data it is clear that some DNA elements, such as antibiotic-resistance genes, can still propagate through large ecosystems (Pruden et al., 2012). As antibiotic-resistance genes are commonly used as markers during plasmid construction, there is therefore major concern that their presence in environmentally released GMMs could contribute to the generation of antibiotic-resistant ‘superbugs’ (Mulvey & Simor, 2009). As antibiotic resistance serves no purpose for the intended function of most GMMs (the genes are merely a legacy of construction), a primary design consideration for synthetic microbes for real-world applications should be not to incorporate antibiotic-resistance genes unless truly necessary.

General design considerations for biosafety

To avoid the use of antibiotic markers and to limit HGT from GMMs, it is tempting to consider abandoning the use of plasmids as vectors for synthetic DNA and disregarding bacterial cells as suitable hosts. Model bacteria such as Escherichia coli and Bacillus subtilis are, however, the cells we understand the most and have had the greatest successes in engineering so far. State-of-the-art applications of synthetic biology are routinely bacterial and almost always their encoding DNA is held on plasmids (e.g. E. coli) rather than introduced into the genome (e.g. B. subtilis) (Khalil & Collins, 2010). This is because plasmid manipulation and iteration is well described and simpler than genomic editing, while also offering larger gene dosage effects due to multiple copies per cell. Plasmids are modular, quick to design and build and, in the absence of selection, can be discarded by the host cell (Silva-Rocha et al., 2013). This latter undervalued biosafety advantage contrasts with directly integrating engineered sequences into genomes, as such constructs are long-lasting without selection and therefore have a greater potential for ‘genetic pollution’, i.e. a long-term presence of unnatural genes, innocuous or otherwise, in the wider environment. This review therefore focuses on plasmid-based systems, although the points herein are generally applicable to a genomic integration strategy.

So what design considerations are of importance to maximize the safety of a plasmid system and in particular to prevent HGT of this naturally mobile element? A clear requirement is to ensure the absence of mobilization genes or origin sequences involved in conjugation or transduction (Davison, 1999). Furthermore, optimization of the vector backbone to minimize homology with mobile elements or the host genome is important in order to discourage sequence recombination (Bensasson et al., 2004). Although these measures will alleviate the predominant sources of HGT amongst microbes, uptake of plasmid by indigenous organisms will still be possible through natural transformation (Thomas & Nielsen, 2005). To further guard against this we describe below a variety of the safety mechanisms and genetic devices that can be used to link a plasmid exclusively to its intended host. These typically require small changes to the host genome, as well as gross changes to the plasmid vector (illustrated by several examples in Fig. 1). The current consensus in the synthetic biology research community is that multiple biosafety mechanisms will be needed to ensure system redundancy in case of component inactivation (Presidential Commission for the Study of Bioethical Issues, 2010; see ). However, the higher the complexity of a safety device, the more prone it may be to disturbance and failure. It is therefore important to understand the expression ‘cost’ of each component, as several in tandem will place an undesirable physiological burden on the host (Glick, 1995) and in turn act as a selective pressure to eject the system (Benner & Sismour, 2005). Counter intuitively, a lack of evolutionary stability of synthetic DNA can serve as a biosafety benefit over time, as system rejection can lead to a pseudo-restoration of a synthetic microbe to a near wild-type state. Clearly, evolutionary pressures are a crucial consideration in design.

While this review focuses solely on the biological mechanisms that can be used to contain GMMs, it should be noted that physical containment is also a powerful biosafety approach. Physical containment is the principal means that currently allows real-world handling of GMMs, either by confining cells to bioreactors (e.g. for biosynthesis applications) or using other methods such as cell microencapsulation (e.g. in alginate or silica beads) (Chang & Prakash, 2001; Nassif et al., 2002; Papi et al., 2005). Biology can achieve a lot in a contained environment; however, physical containment alone offers no guarantees. For example, no matter how ingenious a protective device or material may be for a GMM field application, an inventive way will eventually be found by an operator to compromise it. Failure in this case is a matter of when, not if. Although some form of physical containment is obviously prudent, inbuilt biological mechanisms remain crucial to biosafety.

Dependency devices to reduce the probability of GMM proliferation and HGT

A useful selection of simple genetic systems that could be employed for biological containment is described in Table 1. These are either natural systems that could be reformatted for use in synthetic biology (e.g. toxin–antitoxin pairs or auxotrophies) or existing engineered devices developed previously by others. By splitting the genetic material so that essential components are expressed in trans from both the plasmid and host genome, these systems become dependency devices that either make host viability dependent on maintaining the plasmid or, conversely, ensure that plasmid propagation is dependent upon staying in a specified host.

For toxin–antitoxin pairs, the activity of a small toxin (<15 kDa) is abrogated by a short-lived, cis-encoded antitoxin (Hayes & Van Melderen, 2011; Yamaguchi et al., 2011). In the type I class, antisense RNA inhibits toxin translation; in type II, the antitoxin is proteinaceous (typically <10 kDa); and in type III systems, the antitoxin is an RNA that directly binds and inhibits the toxin. A large number of type I and type II systems are known, and a database of various type II systems is available (Shao et al., 2011). Alongside classic toxin–antitoxin pairs, a variety of other systems can be viewed as toxins with countering antitoxins. Restriction endonucleases are often viewed as the immune system of bacteria, being DNA-cutting toxins whose activity is blocked by immunity-providing DNA methylases (Roberts et al., 2007). Likewise, bacteriocins (Hammami et al., 2010) and bacteriophage lytic systems (Catalão et al., 2013) are also toxic, with the activity of each being suppressed by their own cis-encoded antidote.

First-generation dependency devices consisted of a toxin alone placed under the control of a repressible promoter as the output of a genetic circuit (Molin et al., 1993). Upon repressor removal by the addition of an inducer [e.g. IPTG (Bej et al., 1988)], or by depletion of a targeted contaminant (e.g. 3-methylbenzoate) or key metabolite (e.g. phosphate), toxin expression commences and host death follows (Contreras et al., 1991; Schweder et al., 1992). Unfortunately, this type of system, often called a ‘kill switch’, is prone to low but non-negligible rates of failure (Molin et al., 1993). Random mutation of the constantly repressed toxin gene can lead to toxin inactivation and hence mutant outgrowth when the kill switch is thrown. The probability of such escape has been estimated at 1×10⁻⁶ or greater per cell generation in E. coli when tested in small batch cultures (Knudsen & Karlström, 1991; Moe-Behrens et al., 2013). The addition of a second toxin should theoretically decrease this failure rate to 1×10⁻¹², but survival frequencies of 1×10⁻⁸ per cell generation are more the norm and are independent of toxin type (Knudsen et al., 1995; Pecota et al., 1997). This is due to there being no requirement for inactivating mutations to happen simultaneously; sequential inactivation of each toxin over time is sufficient.

Second-generation devices make toxin expression constitutive, with expression of an antitoxin being the controlled output (Paul et al., 2005; Peubez et al., 2010). This arrangement may improve evolutionary stability: a recent comparison of E. coli genomes found that highly expressed genes are less prone to mutation (Martincorena et al., 2012). However, in studies so far, mutations in second-generation devices still occur at frequencies similar to those for first-generation devices (Pecota et al., 1997; Pasotti et al., 2011). Splitting the pair so that the antitoxin resides on the host chromosome while the toxin is plasmid-encoded does not abate this (Torres et al., 2003), but has the added benefit that, if HGT of a plasmid occurs, a recipient organism may find itself expressing toxin without the antidote (as in Fig. 1). In this arrangement, the probability of biocontainment failure is reduced, as it becomes the product of the toxin inactivation rate multiplied by the frequency of plasmid uptake by a wild-type cell. Third-generation devices may involve coupling synthetic counting circuits to the induction of toxin components (Friedland et al., 2009; Callura et al., 2010). Although intriguing as a route to creating microbes that commit suicide after a defined environmental retention time, this elegant approach would still suffer from the same drawbacks as described above. Regardless of the intricacy of the circuit design, when a kill switch needs to be activated, a lack of selection against mutations in its regulatory or coding sequences can lead to microbial escape. Lastly, it has recently become clear that chromosomally encoded toxin–antitoxin systems play a role in persister formation (Gerdes & Maisonneuve, 2012), in which a fraction of cells survive physicochemical insult (Lewis, 2010). Care should therefore be taken with re-engineered toxin–antitoxin systems that they do not promote this effect and thus negate efforts to induce the elimination of a GMM.

As the examples above indicate, dependency devices based solely on toxins seem destined for failure due to their inability to withstand mutation over time (Schmidt & de Lorenzo, 2012). Indeed, losing genetic information is a common problem for some synthetic biology circuits (Sleight et al., 2010). A more robust approach relies on complementation of deleted or mutated chromosomal genes as a plasmid selection system, i.e. auxotrophy. Mutations that overcome auxotrophic selection in bacteria are unlikely as it is very difficult for a microbe to quickly evolve to reacquire a lost gene's function (Benner & Sismour, 2005). For this reason it should be noted that merely mutating a chromosomal gene promoter to be inactive, rather than entirely deleting the gene, allows for the possibility of reversion mutations (Cranenburgh et al., 2001; Pfaffenzeller et al., 2006a). Auxotrophic selection is also preferential for mitigating the potential harm of successful HGT, as an auxotrophic marker (e.g. a biosynthesis gene) is unlikely to provide any evolutionary benefit to a receiving cell, which is likely to already possess it.

Examples of several auxotrophies are outlined in Table 1 and a list of commonly used auxotrophic bacterial strains can be found elsewhere (e.g. ). For E. coli, the Keio collection of single-gene knockouts indicates potential targets for exploitation (Baba et al., 2006; Yamamoto et al., 2009). While amino acid auxotrophy is usually used, other auxotrophies involving genes in carbohydrate and lipid metabolism are also worth consideration (Baba et al., 2006). For knockouts unable to grow in standard rich growth media (e.g. LB), thiL (cofactor biosynthesis), dapA (peptidoglycan biosynthesis) and thyA (nucleotide biosynthesis) are attractive targets, as supplementation with relatively inexpensive thiamine pyrophosphate, diaminopimelic acid and thymidine (or thymine), respectively can restore growth (Imamura & Nakayama, 1982; Acord & Masters, 2004; Wong et al., 2005).

Auxotrophic selection does, however, suffer drawbacks. Expression levels of the plasmid-borne complementation gene need to be optimized, as overexpression can lead to toxic effects (Vidal et al., 2008). Laboratory-based auxotrophies also typically rely on defined minimal media that lack the key natural metabolite; however, in deployment beyond a lab, heterogeneous environments may remove this selection pressure. This is also exhibited during metabolic cross-feeding, where plasmid-free cells are able to parasitically rely on key metabolite supply from neighbouring prototrophs. Alternative systems impervious to cross-feeding exist but their host strains are difficult to culture prior to plasmid introduction, as their supplements cannot enter the cells (Hägg et al., 2004). Such an approach could still prove useful if used as a final construction step for a pre-optimized synthetic biology system.

Another method of introducing auxotrophies is the use of the amber suppressor system (Kleina et al., 1990). Traditional use involves the introduction of a single amber stop codon (UAG) into an auxotrophy gene, ensuring premature termination of translation unless supplemented with an aminoacyl-charged tRNA carrying the requisite anticodon (CUA). Although the end circuit is still susceptible to interference by metabolic cross-feeding as described above, non-canonical amino acid (ncAA) systems can be utilised in tandem (Hoesl & Budisa, 2012). For example, a mutant aminoacyl-tRNA synthetase–tRNA pair was developed in E. coli to only recognize the ncAA O-methyl-l-tyrosine and exclude any natural amino acid (Wang et al., 2001). When incorporated into amber suppression, host growth only occurs when this ncAA is supplied. Such a system is therefore ‘orthogonal’ to the host translation machinery in that the two are mutually independent (Liu & Schultz, 2010). Cell growth becomes dependent on a synthetic metabolite being provided, thus allowing for control of cell proliferation. Given, however, that E. coli uses the amber codon for termination in over 300 ORFs (Blattner et al., 1997), this system is far from ideal. Inappropriate read-through of native genes can lead to deleterious effects and amber suppression is never fully effective in E. coli. Recent work on engineering modified cells for amber suppression has alleviated some of these problems, allowing read-through of several amber codons in a single gene (Hoesl & Budisa, 2012). The use of amber codons within a synthetic DNA cargo is also an attractive mechanism against HGT, as translation of such a gene would be prematurely halted in wild-type cells. A potential issue, however, is that the yield from translation of this cargo may be diminished due to inefficiencies in amber suppression, perhaps to the extent that the circuit is no longer fit-for-purpose.

While loss-of-function auxotrophies are evolutionarily hardier than toxin–antitoxin systems, they do not prevent a plasmid establishing itself in a wild-type microbe, especially if the synthetic DNA ‘cargo’ of the vector provides an evolutionary advantage. A further way of enforcing plasmid biocontainment is to make its replication dependent on a specific host using a system known as conditional origin of replication [COR, (Soubrier et al., 1999)]. Conditional plasmid origins use a cis-encoded replication initiation protein (del Solar et al., 1998), which, if relocated to the host chromosome, can still perform its function in trans in modified cells (Kittleson et al., 2011). In this split replication machinery scenario, any uptake of a COR plasmid by a wild-type microbe would only be transient due to an absence of the requisite replication initiation protein.

Some of the devices described above have been used in tandem. A gene therapy vector developed in 1999 used a dual-dependency device where an amber suppressor tRNA gene and a COR (R6K ori-γ) were supplied on a plasmid transformed into E. coli (Soubrier et al., 1999). The host was genomically modified to contain the R6K plasmid replication initiator (π), and the argE gene was mutated to include an amber codon near the beginning of its ORF. In the absence of the plasmid-supplied suppressor tRNA, the GMM was an arginine auxotroph in minimal media and the plasmid itself was unable to replicate in wild-type E. coli. Ramos and colleagues also updated the aforementioned 3-methylbenzoate-responsive kill switch (Contreras et al., 1991) so that this compound repressed expression of a plasmid-bound toxin (HokC) as before, but additionally induced expression of a modified genomic copy of asd, necessary for peptidoglycan synthesis (Ronchel & Ramos, 2001). During experiments in 3-methylbenzoate-contaminated soil their modified Pseudomonas putida strain survived and maintained its plasmid. Upon contaminant depletion, however, this GMM dropped below detectable levels after 25 days. Mutant escape was below their detection limit and was therefore estimated to occur at a probability of 1×10⁻⁹ or less per cell generation.

DNA barcodes to trace synthetic biology designs

Assessing the efficacy of biosafety systems by measuring GMM spread and HGT events in sample environments is a complex task. However, with DNA sequencing becoming rapid and affordable, direct sequencing of environmental samples can now be used to identify contaminating synthetic DNA. Synthetic operons can be designed to contain genetic ‘barcodes’, which, if indexed to a pre-release database, could be used to identify their origin and particulars. Others have embedded ‘DNA watermarks’ in multiple genomic locations to aid in identifying their engineered cells (Gibson et al., 2010). Barcodes not only aid in identifying GMMs in the environment, but can be used commercially to mark proprietary strains that may be stolen during industrial espionage. To guard against stolen strains simply having their DNA recoded (removing evidence of theft), cryptography approaches have also been applied that introduce cryptic ‘DNA watermarks’ either into multiple genomic locations in engineered cells (Gibson et al., 2010) or directly into synthetic genes via manipulation of their codon usage (Liss et al., 2012). This latter approach, presuming such codon changes are functionally neutral, would be especially suitable to exploit for tracing synthetic DNA in the environment. Embedding watermarks directly within genes likely to experience positive selection in the environment will help ensure their incorporation during instances of HGT, whereas upstream or downstream barcode elements may be lost during recombination events (Thomas & Nielsen, 2005).

Orthogonal systems for semantic containment

While the biosafety systems outlined above can presently be incorporated into synthetic biology designs, more radical solutions to containment are on the horizon. Researchers are now pursuing biosafety through semantic containment, whereby a ‘genetic firewall’ is erected between synthetic microbes and natural organisms much like a linguistic barrier (Schmidt, 2010). This process involves refactoring, in which the composition or order of the basic genetic material of a GMM is changed without altering its encoded output (i.e. polypeptide sequence). In this manner a synthetic gene may no longer be meaningfully read by a natural organism. Such orthogonality is possible in a number of ways and has been recently reviewed by Schmidt & de Lorenzo (2012); highlights, including subsequent work, are briefly mentioned below. One approach involves refactoring a cell’s codon usage, and has now led to the substitution of all 314 amber codons in the E. coli genome with an alternative stop codon (Wang et al., 2009; Isaacs et al., 2011). In these cells, the liberated UAG codon is freed to encode ncAA auxotrophies at will (Chin, 2012; Church lab, unpublished data). Genome modification en masse could also be conceivably used to shuffle codon assignments in a synthetic microbial genome, resulting in the same protein products as wild-type cells but via a different genetic code. Translation of an altered synthetic gene from such a system by a natural organism would give an effectively mistranslated product. Evolved ribosomes, which recognize non-natural ribosome-binding sites for translation or translate recognizing a quadruple-base-pair code, are another way of obtaining genetic code orthogonality (Neumann et al., 2010). While both of these orthogonal approaches utilize natural nucleic acids, others are pursuing synthetic versions (Kwok, 2012). An E. coli incorporating 5-chlorouracil into its DNA was recently created by weaning cells off the thymine nucleotide over 25 weeks (Marlière et al., 2011). At the end of this directed evolution, descendant microbes grew only in the presence of the synthetic nucleotide. Beyond this, efforts to expand the genetic code beyond four bases are advancing. Alternative base pair combinations accepted by natural DNA polymerases in vitro have been found (Leconte et al., 2008; Yang et al., 2011), and work to prove these synthetic bases work with plasmids in vivo is ongoing. Xeno nucleic acids (XNA), where the backbone sugars of DNA are changed, seem more problematic as natural DNA and RNA polymerases do not recognize them. Although this currently limits their utility in vivo, recent in vitro work has taken a first step towards solving this problem by creating polymerase mutants that can use DNA as a template for XNA synthesis and vice-versa (Pinheiro et al., 2012). The use of XNA in vivo is, however, many years away. XNA to XNA replication needs to be established, and a xenobiotic host would require re-engineered RNA polymerases, as well as other XNA-compatible replication and transcription components (Herdewijn & Marlière, 2009; Schmidt, 2010).

The above approaches could lead to effective semantic containment within decades; however, this would not stop a refactored microbe from competing at the physiological level with natural flora and fauna during environmental release. As per Asilomar, short-lived microbes should be utilized (Berg et al., 1975). For alternative base pair and XNA systems, this is easily achievable without requiring attenuating knockouts, as the requirement for exogenous synthetic components would make them auxotrophic by definition. Supplementation at the site of release with the required synthetic compound would be required, and xeno-synthetic microbial death would ensue upon xeno-metabolite withdrawal. To stringently guard against evolution around this auxotrophy (i.e. removing synthetic-compound dependence), the xeno-metabolite would be at least two steps of synthesis away from any natural compound (Schmidt, 2010). Although only theoretical at this stage, such a system should represent the safest biocontainment mechanism possible through the incorporation of both trophic and semantic containment (Marlière, 2009).

Now and the future

Orthogonal biological systems and xenobiology offer significant hope for microbial cells designed to have minimal genetic interaction with nature, and further development of these will surely proceed. However, while truly orthogonal, environmentally relevant synthetic microbes remain years away, the repurposing of natural components remains our best arsenal for inbuilt biological safety. With no single perfect mechanism, the current consensus is that the bare minimum of safety for a deployed GMM should consist of multiple devices of different types (Presidential Commission for the Study of Bioethical Issues, 2010). This redundancy would present a GMM with several evolutionary hurdles to overcome simultaneously in order for system failure to occur, therefore greatly safeguarding against ‘life finding a way’ (Benner & Sismour, 2005). Physical containment should be used where suitable, and microencapsulation systems already exist that can be implemented (Chang & Prakash, 2001; Nassif et al., 2002). As it is prudent not to use antibiotic-resistance markers, auxotrophic selection appears wise to incorporate into design, although care must be taken that the auxotrophy chosen is appropriate for the environment into which the GMM will be applied. DNA barcodes or watermarks provide an efficient route for tracing synthetic microbes, and plasmids with conditional origins of replication can guard against plasmid establishment following instances of HGT. The use of toxin–antitoxin pairs to secure plasmids to hosts and vice-versa is more problematic due to a lack of evolutionary stability.

A further important mechanism that can be implemented is imperfect retention, i.e. cells that survive for months but not years in the environment, or plasmid-based constructs that are gradually lost after their hosts are deployed. Asilomar-recommended attenuated microbial strains and those deficient in cell maintenance (Schweder et al., 1995) so far seem unable to establish themselves in tested environments (Benner & Sismour, 2005; Cases & de Lorenzo, 2005; de Lorenzo, 2010). As short-term environmental retention times are more palatable (Church, 2005), it seems prudent for synthetic biology to be intentionally designing GMMs with half-lives of days or weeks where the intended application permits. Analysis of evolution tells us that losing a genetic circuit is much easier than obtaining it anew (Benner & Sismour, 2005). As most engineered cells are being made to perform work superfluous to their critical functions, they are likely to be out-competed beyond the comforts of the lab and therefore either die out or eject their synthetic circuits.

Further thought is required on how to design synthetic constructs and microbes to be intentionally out-competed over time. For this research to progress, more quantitative data are needed for how GMMs perform in sample environments. The current lack of in-depth testing means that it is hard to accurately assess which safety mechanisms and designs are best at preventing ecological invasion and HGT. Only initial studies with toxin systems have so far been informative, showing that simple kill switches alone are not adequate. Priority should be given to future studies assessing the successes and failures of various containment mechanism combinations in appropriate environmental situations. If a successful case study were to be taken out of the lab and put into practice, it would greatly inform the field.

Whether any GMM application will be approved in the near future for real-world use outside controlled premises is not clear. Realistically, if risk assessment primarily hinges on what effect a synthetic gene may have if accidentally established in the wild, then mitigating biosafety mechanisms are a secondary consideration (Molin et al., 1993). However, the various regulatory bodies operating in biosecurity, healthcare, agriculture, etc. take differing paths to their decisions (Rodemeyer, 2009) and may show a variety of sensitivities to the use of cisgenic, transgenic or syngenic material (i.e. potential subclasses of GMMs). Scientists must realize that for some proposed real-world synthetic biology applications, the benefits of their deployment may never outweigh the perceived risks, which range from genetic pollution via HGT of innocuous synthetic genes through to the dual-use of technologies by those intent on causing harm (Dana et al., 2012; Hoffman et al., 2012). In working towards future applications of GMMs, researchers therefore should not only aim to incorporate biosafety mechanisms into their designs to help alleviate potential risks, but should also seek to engage stakeholders and regulators, who will ultimately decide how safe is safe-enough (Bhattachary et al., 2010; Presidential Commission for the Study of Bioethical Issues, 2010).

Encouragingly, the synthetic biologists of the future are already showing a serious interest in interacting with society beyond the lab and also in incorporating biosafety considerations into their designs. In the undergraduate synthetic biology competition iGEM (), the Imperial College 2011 team discussed their project design with environmental scientists and incorporated a ‘Gene Guard’ device to minimize HGT by using the T4 bacteriophage holin/endolysin system in trans. Towards the same ends, the Paris Bettencourt 2012 team created ‘bWARE’, a system combining physical encapsulation, an endonuclease/bacteriocin-dependency device and semantic containment via an amber suppressor system. They also compiled information on the safety mechanisms attempted by other teams (available at ), which will serve as a useful compendium for others.

Ultimately, the ideal safety strategy for a particular GMM depends on not only biosafety mechanisms but also the genetic cargo the cell contains, the task this performs, the intended end-user and the environment in which it will be used. While biosafety is an under-reported aspect of synthetic biology research, it is clear that it needs to be better addressed if the full potential of synthetic biology is to be realized.