Controlling the implementation of transgenic microbes: are we ready for what synthetic biology has to offer?

Summary

Synthetic biology has promised and delivered on an impressive array of applications based on genetically modified microorganisms. While novel biotechnology undoubtedly offers benefits, like all new technology, precautions should be considered during implementation to reduce the risk of both known and unknown adverse effects. To achieve containment of transgenic microorganisms, confidence to a near scientific certainty that they cannot transfer their transgenic genes to other organisms, and that they cannot survive to propagate in unintended environments, is a priority. Here we present an in-depth summary of biological containment systems for micro-organisms published to date, including the production of a genetic firewall through genome recoding and physical containment of microbes using auxotrophies, regulation of essential genes and expression of toxic genes. The level of containment required to consider a transgenic organism suitable for deployment is discussed, as well as standards of practice for developing new containment systems.

Introduction

1.1. History of containment

Microorganisms have been used inadvertently in the processes behind food production for thousands of years, with various bacterial and fungal species playing key roles in brewing beer and the fermenting of wine, the baking of bread, and the process of turning milk to yoghurt. However, microbiology as a scientific discipline did not solidify until the 17^th century, with the exploration of microorganisms as the possible vector for human and animal diseases. It was with these observations that the motivation to control and contain bacterial growth became apparent. Medical advances attributed to sterile and abiotic techniques played an integral part in the control of pathogens. In the 20^th century, the discovery and use of antibiotics lead to a revolution in bacterial containment by the application of small molecules.

Today we use microbes for an astonishing range of functions. In addition to the aforementioned production of ethanol, CO₂ and lactic acid for alcoholic drinks, bread and yoghurt respectively, both wild-type and engineered microbial species are used to produce a variety of industrially relevant products. Examples include; enzymes such as amylase (Sundarram & Murthy, 2014; Yoneda, 1980) and proteases (Razzaq et al., 2019); polysaccharides for use in food, cosmetics, and pharmaceutical products such as xanthan gum (Santos et al., 2000) and dextran (Sarwat et al., 2008); nutrients in the form of amino acids, nucleotides, vitamins and organic acids (Adrio & Demain, 2010); and pharmaceuticals themselves including insulin (Baeshen et al., 2014), chemotherapeutics (Łukasiewicz & Fol, 2018) and antibiotics (Clardy et al., 2009).

In addition to their use as biological factories, the advent of synthetic biology has allowed microbes to be manipulated by humans to facilitate desirable processes, such as the bio-remediation of polluted environments and for use as living therapeutics. Microbial bio-remediation generally involves the breakdown of an undesirable substrate, rather than concentrating on the production of a functional product. Applications include sewage treatment (Dhall et al., 2012), removal of pesticides (Paridah et al., 2016) and aromatic compounds (McClure & Venables, 1986) from soil, and clean-up of oil spills (Mapelli et al., 2017). Although microbes have been consciously used as therapeutics for over a century (Coley, 1893), the last decade has seen a plethora of possible applications presented. A non-exhaustive list includes diagnostic tools (Kotula et al., 2014; Riglar et al., 2017) and treatments for the human digestive system (Braat et al., 2006), a treatment for oral mucositis (Caluwaerts et al., 2010), HIV prevention (Lagenaur et al., 2011) and cancer immunotherapy (Zheng et al., 2017a). Various reviews have covered the novel applications that synthetic biology allows for bioremediation (de Lorenzo, 2009; Pieper & Reineke, 2000; Sayler & Ripp, 2000) and for living therapeutics (Riglar & Silver, 2018).

Despite the broad acceptance that there has been no known instance of a transgenic microbe being conferred a fitness advantage in the wild, frequent release of transgenic microbes into real world environments presents the possibility of such an individual manifesting. The responsibility of international governing bodies to understand and mitigate this potential problem has been established and ratified in legal treaties such as the Cartagena Protocol and the Rio Declaration on Environment and Development. Transgenic microbes capable of proliferating outside of a lab would lead to their uncontrolled propagation and persistence in an environment that may affect biological processes through interaction with indigenous populations, therefore reducing biodiversity and disrupting food webs. Increases in the risk of transfer of transgenic DNA through horizontal gene transfer could, in principle, lead to novel pathogens. (NIH FAQs, 2019; Risks SCoEaNIH, 2015).

Although the full risks associated with the spread of transgenic microbes are not known, it is this very uncertainty that makes it important to implement control and regulation over transgenic microbial growth. Proving absolutely that a particular transgenic microbe capable of proliferation in the wild poses no environmental risk upon release is inherently an effort in futility, as biological processes contain too many unknown interactions for a perfectly predictive model to be formed. In cases where such uncertainty is prevalent, the precautionary principle can be applied to help form guidelines for the implementation of such a novel technology (A. Stirling, 2007), with critics agreeing it can have a positive effect in such a situation (Peterson, 2007). For example, if a strain is developed capable of treating a Salmonella infection (Riglar et al., 2017), it is prudent to introduce biological safeguards that would reduce the possibility of such a microbe escaping its intended environment (F. Stirling et al., 2017) even if we cannot fully predict what, if any, negative consequences may arise from such an escape. It is our view that transgenic microbes offer a range of solutions to environmental and health related issues, and in the coming decades they should and probably will be implemented. To facilitate this, exercising rigorous precaution will both increase public trust (A. Stirling et al., 2018) and, most importantly, reduce the possibility of negative consequences.

For these reasons, the field of biocontainment has been of interest for decades. The Asilomar conference (Berg et al., 1975) laid out guidelines for the introduction of auxotrophies to prevent escape of transgenic microbes. With the advent of synthetic biology, it has been judged necessary to repeatedly revisit the topic, with a number of reviews appearing in the intervening years (Chari & Church, 2017; Diwo & Budisa, 2019; Lee et al., 2018; Moe-Behrens et al., 2013; Molin et al., 1993; Schmidt & De Lorenzo, 2012; Torres et al., 2016; Wright et al., 2013). The applications of transgenic microbes covered so far can broadly be divided into two categories; applications where the entire intended purpose can be carried out in a controlled setting, and those where the intended purpose is designed to take place in an uncontrolled environment. Microbes engineered for production predominantly fall into the former category, whereas the majority of the second category are those microbes engineered to facilitate a process. However, both categories face the same requirements, that i) that transgenic material is not spread to other species and ii) transgenic microbes are only viable in their intended environment. Containment systems to date predominantly address one or both of these two issues.

A key means to quantify the effectiveness of a containment system used throughout this review is to measure the ‘survival ratio’ it imparts on a population. The survival ratio is the term used to define the ratio of colony forming units (cfu) in non-permissible conditions to the cfu in permissible conditions. For a survival ratio of 10⁻⁴, a population of 10,000 bacteria would yield only 1 survivor on transition to non-permissible conditions. Although not all systems can be quantified in such a manner - such as the essentializer in (F. Stirling et al., 2017) - it is a useful tool for comparing different approaches and when assaying a systems effectiveness. In the United States, the National Institutes of Health has recommended a guideline providing that the “escape of the recombinant or synthetic nucleic acid molecule either via survival of the organisms or via transmission of the recombinant or synthetic nucleic acid molecule to other organisms should be less than 1 in 10⁸ under specified conditions” (NIH FAQs, 2019, Apendix 1–1-B). A survival ratio of 10⁻⁸ would currently meet these requirements. For many applications, it is likely that this threshold is too high, which is discussed further in section 1.5.

1.2. Preventing the spread of transgenes

Every transgenic organism constitutes an example of one or more units of genetic material existing in an organism or in an arrangement that does not exist in nature. This enables interactions and transfers of genetic material that have not previously been possible. The various strata of life display both unique and interconnected mechanisms for spreading genetic material, and the intentional transfer of such material by humans is not only possible, but performed on a routine basis. The potential negative consequences of this can often be categorised as low risk, like the expression of the green fluorescent protein (GFP) from the jellyfish Aequorea victoria. The use of GFP is considered harmless because no adverse consequences have been observed from its use in transgenic strains (Lee et al., 2018; Moe-Behrens et al., 2013), and hypothetical problems that have been conceived of should transgenic strains expressing it be released are mild or uncertain. However this by no means indicates that if GFP were spread throughout an environment it did not originate from, that it is impossible for it to have a negative impact. Since it would be essentially impossible to prove the innocuousness of the spread of a transgene in all conceivable scenarios, a more feasible and assured approach is to control its inability to spread in the first place with a reasonable level of confidence.

Early attempts at preventing horizontal transfer of genetic material between microbes relied upon toxin-antitoxin systems expressed on plasmids and genomes (Diaz et al., 1994; Torres et al., 2000; Diaz et al., 2003) or host repression of a plasmid borne toxin (Knudsen & Karlstrom, 1991) to prevent the spread of transgenic plasmids to other microbes. A similar approach has been proposed that uses origins of replication that are split between a plasmid and the host genome to prevent plasmid spreading (Wright et al., 2013). These two techniques were combined (Wright et al., 2015), achieving a survival ratio of <10⁻³ for each method individually but the study did not report the survival ratio from the combined strain. These systems are summarized in Table 1. It is of note that the stability (see section 1.4) of these systems, ie. the capacity of the system to maintain its function when passaged in permissible conditions, is not reported for any system.

Table 1.

Containment systems designed to prevent the spread of transgenic plasmids.

Year	Lead Author	Species	Conditional Origin	toxin/antitoxin system	Survival ratio	Stability
1991	SM. Knudsen	E. coli		RelF	10^−5	not tested
1994	E. Diaz	E. coli		colicin E3	10^−4	not tested
2000	B.Torres	E. coli		EcoRI	10^−4	not tested
2003	B.Torres	E. coli		EcoRI, colicin E3	10^−8	not tested
2014	O. Wright	E. coli	ColE2	Kid or ζ	<10−3	not tested

To achieve a ‘hard lock’ (where the probability of the transfer and expression of transgenic material can be deemed low enough as to be negligible) it is necessary to address the ‘central dogma’ of biology and unpick some of its core truths. Currently, all natural organisms are based around the same genetic code. The four deoxyribose nucleotides are transcribed into the four ribose nucleotides, which are in turn translated using the ubiquitous triplet codon code into the twenty standard amino acids that make up all proteins. Several studies have sought to alter this dynamic in one way or another. By engineering a tRNA synthetase, it has been shown that a quadruplet codon scheme can be implemented in order to incorporate non-natural amino acids (Chatterjee et al., 2014; Hankore et al., 2019). In addition Escherichia coli has been engineered to survive on an artificial nucleotide (Marlière et al., 2011) or to incorporate non-natural amino acid pairs (Malyshev et al., 2014).

For containment applications, the most effective modification to the fundamental tenets of life upheld in the central dogma is to recode a genome so completely that it no longer requires the tRNA machinery for a specific codon or set of codons. This was first accomplished in bacteria with E. coli by removing all instances of the TAG stop codon and its corresponding tRNA (Lajoie et al., 2013), and again more recently removing TAG along with the serine encoding codons TCG and TCA (Fredens et al., 2019). Alternative methods have been published for the complete recoding of an organism such as an integrase based approach to achieve a 57 codon E. coli genome (Ostrov et al., 2016), a technique termed SIRCAS (Step-wise Integration of Rolling Circle Amplified Segments) used to remove two codons from the genome of Salmonella typhimurium (Lau et al., 2017) and an extensive international effort in Saccharomyces cerevisiae that successfully removed all instances of the TAG stop codon (Mitchell et al., 2017; Richardson et al., 2017). The different recoding approaches are summarised in Table 2 and in previous reviews (J. Kuo et al., 2018).

Table 2.

Recoding Approaches Attempted to Date

Year	Lead Author	Species	% Recoded	Recoding Method	Codons Removed	Amino Acid
2011	F. Isaacs	E. coli	100	MAGE/CAGE	TAG	Stop
2016	N. Ostrov	E. coli	100a	integrase-based segments approach	AGA, AGG, AGC, AGT, TTA, TTG, TAG	Arg
						Ser
						Leu
						Stop
2017	Y.H. Lau	S. typhimurium	4.5	SIRCAS	TTA, TTG	Leu
2017	S. Richardsonb	S. cerevisiae	100	SWaP-In	TAG	Stop
2019	J. Fredens	E. coli	100	REXER	TCG, TCA, TAG	Ser
						Stop

^aThis was achieved over 87 different strains.

^bThe recoding of S. cerevisiae was a large collaborative effort that resulted in 7 publications in a special issue of Science (March 10, 2017), of which S. Richardson is one lead author.

Freeing up a codon, and subsequently its associated tRNA, allows for four novel containment possibilities; 1) A recoded organism could have transgenic genes that are only translated in a functional manner by the host by inserting a recoded stop codon throughout the open reading frame, thus initiating early termination of translation in any organism not recoded in the same manner (Figure 1A). It should be noted this approach cannot prevent the transfer of transgenic DNA and subsequent mutations that remove recoded codons would allow expression. 2) A second approach could be to incorporate a broad range highly lethal toxin, such as an endonuclease, alongside any transgenic circuit. Such a toxin could be prevented from expression in the transgenic strain by incorporating canonical codon usage. Upon transfer to a wild-type strain, toxin expression would cause cell death. As transformation and transduction of genomic material rely upon homologous recombination, transgenic circuits would reliably be transferred in single units (Figure 1A). This approach prevents the transfer of the transgenic material in the first place, but has the drawback of a single mutation disabling the toxin in the transgenic strain would prevent containment system functioning. 3) A recoded organism without the full contingent of tRNA’s and their accompanying release factors is unable to translate most novel genetic material it acquires, blocking a potential avenue of evolution. (Figure 1B). 4) Novel tRNA synthetases that incorporate a non-natural amino acid can be used alongside an unassigned codon to engineer an auxotrophy, forming an effective containment system (Mandell et al., 2015; Rovner et al., 2015), explored further in section 1.3 (Figure 1C).

Figure 1: Schematic of biological containment mechanisms.

A) Methods for addressing transfer and expression of genetic material in unintended hosts B) methods of preventing DNA from other bacteria being expressed in a transgenic strain C) methods for preventing the physical translocation of bacteria from their intended environment to an unintended environment, and D) methods for removing bacteria from their intended environment if they are no longer required. Physical forms of containment such as barriers and waste management are not included.

A major obstacle to the widespread adoption of genome recoding is the ability to devise recoding schemes for a diverse range of organisms. Published recoding efforts to date have concentrated on E. coli and S. cerevisiae, and to a lesser extent S. typhimurum, some of the most comprehensively studied model organisms. Recoding a genome requires an extensive knowledge of an organism’s essential and overlapping genes, as well as established protocols for engineering and growth. Although the cost and methods of DNA synthesis currently makes recoding on a routine basis difficult, falling costs will make this more viable in the future (James Kuo et al., 2018).

It is important to note that a recoded organism represents a divergence from all previously known forms of life, and while this is not inherently dangerous, additional caution should be employed when considering their biocontainment.

1.3. Containment of transgenic microbes to their intended environment

Applications of transgenic living microbes are undertaken in controlled environments. Here, a controlled environment is defined as one where physical barriers, effective waste management, and enforced regulation can easily be assured, including both laboratory research and industrial production. There are many published containment systems that have been designed to prevent transgenic microbes from escaping their intended environment. These generally fall into one or more of the following categories: allowing for the expression of an essential gene, regulation of a toxic gene, and supplementation of an auxotrophy (Figure 1C). In addition, regulation of a toxic gene has been shown to modulate the self-maintenance of a transgenic strain (Figure 1D). Table 3 compiles a reasonably exhaustive list of 35 containment systems, that take advantage of essential and toxic gene regulation, engineered auxotrophy or a combination of several mechanisms. It includes their mode(s) of containment, the signal(s) they respond to, their survival ratio, the evolutionary stability of the system when grown in permissible conditions and whether they are designed for use in a controlled or uncontrolled environment. Each row of the table represents a single containment system in a single strain. Papers that published multiple systems have generally been represented with whichever system had the lowest survival ratio, unless the modes of regulation are so disparate as to be considered separately of interest. Unless otherwise stated, the following systems are solely applicable to the processes described in Figure 1C. Although containment systems have been published for non-microbial purposes (Deans et al. 2007), they are not reported on in this review.

Table 3.

Containment systems designed to prevent escape of transgenic microbes.

Year	Lead Author	Species	Auxotrophy	Essential gene	Toxic gene	Responds to	Survival ratio	Stability	Used in
1987	S. Molin	E. coli			hok	tryptophan	10−4	not tested	con
1988	A.K. Bej	E. coli			hok	IPTG	<10−6	not tested	con
1991	A. Contreras	E. coli			gef	benzoates	10−6	not tested	un
1991	S.M. Knudsen	E. coli			relF	IPTG	10−8	not tested	con
1992	A.K. Bej	P. putida			gef	IPTG	10−5	not tested	con
1993	G. Recorbet	E. coli			sacB	sucrose	10−3	not tested	con
1994	I. Ahrenholtz	E. coli			nucA	temp. (cI857)	10−5	not tested	un
1995	SM. Knudsen	E. coli			relE	IPTG	10−7	not tested	con
1996	MT. Munthali	P. putida			colE3	3-methyl benzoate	N/A	not tested	un
1997	P. Szafranski	P. putida			streptavidin	3-methyl benzoate	10−7	not tested	un
1998	L. Molina	P. putida			gef	3-methyl benzoate	10−8	not tested	un
2000	P. Kristoffersen	S. cerevisiae			relE	galactose	N/A	not tested	con
2001	MC. Ronchel	P. putida			gef	3-methyl benzoate	<10−9	not tested	un
2003	L. Steidler	L. lactis	thymidine			thymidine	10−7	not tested	un
2005	A. Balan	S. cerevisiae			nucA	glucose	10−5	not tested	con
2008	W. Kong	S. typhimurium		asdA, murA		arabinose	10−4	not tested	con
2009	Q. Li	E. coli			nucA	arabinose	N/A	not tested	con
2010	JM. Callura	E. coli			ccdB/λ lysis/lexA3	aTc, arabinose, IPTG	10−3	not tested	con
2015	CTY. Chan	E. coli		murC	ecoRI	aTc, IPTG	<10−5	Unstable after four days	con
2015	CTY. Chan	E. coli		murC	ecoR1	IPTG, gal, cellobiose	<10−8	Unstable after four days,	con
2015	RR. Gallagher	E. coli	Biotin	ribA, glmS	ecoRI	aTc, IPTG	<10−12	stable for >110 gen.	con
2015	Y. Cai	S. cerevisiae		HHTS, HHFS		galactose, estradiol	<10−10	not tested	con
2015	DJ. Mandel	E. coli	bipA			bipA	<10−11	not tested	con
2015	AJ. Rovner	E. coli	pAcF α			pAcF α, arabinose	<10−11	not tested	con
2015	G. Lopez	E. coli	benzothazole			benzothiazole	10−7	not tested	con
2016	DI. piraner	E. coli			ccdB	temp.	10−5	not tested	un
2016	S. Huang	E. coli		blaM, cat1		cell density	10−5	not tested	con
2016	R. Hirota	E. coli	Phosphite			phosphite	<10−12	not tested	con
2017	N. Agmon	S. cerevisiae		SEC17		estradiol	10−8	not tested	con
2017	F. Stirling	E. coli			ccdB	temp.	10−5	Stable for >140 gen.	un
2017	F. Stirling	E. coli			ccdB	cI and/or Cro	N/A	Stable for >140 gen.	un
2018	RL. Clark	S. sp. PCC7002	CO2			CO2	10−9	not tested	con
2019	F. Stirling	E. coli			doc	pH	10−6	Stable for >100 gen.	un
2019	F. Stirling	E. coli			ccdB/doc	temp., pH	<10−11	Stable for >100 gen.	un
2019	F. Stirling	E. coli			doc	pH2	10−4	not tested	un

Strains: Escherichia coli, Pseudomonas putida, Saccharomyces cerevisiae, Lactococcus lactis, Salmonella typhimurium, Synechococcus sp. PCC7002. Each row represents a single containment system in a single strain. For “Used in” column, con/un refers to controlled or uncontrolled environment.

¹These genes are essential in the presence of the antibiotics carbenicillin and chloramphenicol.

²This containment system only expresses the toxin upon two, non-consecutive exposures to low pH.

Containment systems for a controlled environment

Of the 35 systems, 22 respond to one or more small molecules intended to be provided by human intervention, and are therefore designed to contain the transgenic strains to a controlled environment. Twelve systems consist of a basic design concentrating on a single molecule regulating a single toxic gene (Balan & Schenberg, 2005; Bej et al., 1992; Bej et al., 1988; Knudsen & Karlstrom, 1991; S. Knudsen et al., 1995; Kristoffersen et al., 2000; Li & Wu, 2009; Molin et al., 1987; Recorbet et al., 1993) or one or more essential genes (Agmon et al., 2017; Cai et al., 2015; Kong et al., 2008).

The last decade has seen a progression towards more complex containment systems. Three examples from the Collins lab (Callura et al., 2010; Chan et al., 2015) explore the capabilities of synthetic circuits to respond to multiple inputs controlling the expression of both toxic and essential genes, and are excellent examples of modular and customisable containment systems. Another example comes from the Isaacs lab (Gallagher et al., 2015), who combined an auxotrophy, suppression of two essential genes and the expression of a toxin to construct a containment system with a survival ratio below their detection limit of 10⁻¹².

A unique containment system based on cell density (Huang et al., 2016) depends on collective expression of an antibiotic resistance gene whose expression is stimulated by the quorum sensing factor N-acyl homoserine lactone (AHL). At high cell densities, enough antibiotic resistance is expressed at a population level and released (upon cell lysis, expressed intracellularly, or excreted) to allow individual cells to survive. However if individual cells split off from the “microbial swarmbot”, insufficient resistance proteins would be expressed to allow survival. Although this system currently requires the continued application of antibiotics to provide the toxic function, it is suggested that future iterations could modulate the expression of toxic or essential genes, removing the requirement for supplementation and facilitating its use in an uncontrolled environment.

The final 5 systems designed to be used in a controlled environment are all auxotrophies, categorized as such because they require the uptake of the small molecule they are responding to for some aspect of their metabolism that cannot otherwise be self-synthesised. To construct a system that responded to atmospheric CO₂ (Clark et al., 2018) the CO₂ concentrating mechanism of the cyanobacteria Synechococcus sp. PCC7002 was removed. This only allowed survival when the transgenic strain was grown in an environment with at least 5% ambient CO₂, with atmospheric levels of CO₂ resulting in a survival ratio of 10⁻⁹. Another study engineered an E. coli phosphite auxotroph by removing all phosphate production pathways except via phosphite uptake, achieving a survival ratio below their detection limit of 10⁻¹² (Hirota et al., 2017). A system for making Synthetic auxotroph’s based on a Ligand-Dependent Essential genes (SLiDE) was developed (Lopez & Anderson, 2015). SLiDE was used to develop a strain were the function of three essential genes was dependent upon the presence of the ligand benzothiazole. Finally, two synthetic auxotrophs were designed using the E. coli recoded strain with all instances of the TAG stop codon removed, mentioned in section 1.2 (Lajoie et al., 2013). Novel tRNA machinery was introduced to incorporate non-natural amino acids into the primary protein sequence of several essential genes (Mandell et al., 2015; Rovner et al., 2015). These last two methods both showed extremely robust containment, below their detection limit of 10⁻¹¹.

Containment systems for an uncontrolled environment

The other 13 of the 35 systems in Table 1 were designed for application outside of a controlled environment. A thymidine auxotrophy containment system for Lactococcus lactis engineered to act as a therapeutic for Crohn’s disease when applied to the human digestive system showed a survival ratio of 10⁻⁷ both in vitro and when applied to a porcine model (Steidler et al., 2003). Although unable to propagate outside of a controlled environment, the functional application for this strain of expressing human interleukin-10 did not require DNA replication, allowing its application in an uncontrolled environment.

Five systems were designed for application in microbes engineered to facilitate the bioremediation of soil by the degradation of benzoates (Contreras et al., 1991; Molina et al., 1998; Munthali et al., 1996; Ronchel & Ramos, 2001; Szafranski et al., 1997). While the degradation target, benzoates, are present, toxin expression is repressed allowing survival. Because of this they respond to the presence of small molecule regulators, but will initiate population suicide upon the completion of their task or translocation from their intended location without the need for human intervention (Figure 1C and 1D).

A system was developed that was designed to terminate a bacterial population upon a loss of function mutation, based on the bacteriophage lambda cI/cro regulatory system (Stirling et al., 2017). In the presence of either cI or Cro, the toxin is repressed. However, in the absence of either, repression is relieved and the toxin is expressed. This system allows for the maintenance of functionality of a specific transgenic strain, checking to see if the transgenic element is present and terminating the strain if it is not (Figure 1D).

Three systems that responded exclusively to temperature have been published, each with a different mode of control. The first controls expression of the Serratia marcescens nuclease nucA with the temperature sensitive mutant of cI, cI⁸⁵⁷ (Ahrenholtz et al., 1994). At 42 °C, cI⁸⁵⁷ is unable to function, and the toxin is expressed resulting in a survival ratio of 10⁻⁵. The second controls expression of the DNA gyrase inhibitor ccdB using a mutant tlpA repressor that is activated below 36 °C. Using this system they were able to show the containment of a bacterial population to the mammalian gut with a survival ratio of 10⁻⁵ (Piraner et al., 2016). The third uses the regulatory region from cold shock protein A to control expression of the toxin ccdB, also achieving containment to the mouse gut with a survival ratio of around 10⁻⁵ (Stirling et al., 2017).

Finally, three systems that respond to pH have been reported (F. Stirling et al., 2020). The first controls the expression of the toxin Doc with the pH sensitive promoter P_asr, achieving a survival ratio of 10⁻⁶ when exposed to pH 5 conditions. The second combines the pH sensitive expression of Doc with the temperature sensitive expression of CcdB to achieve a survival ratio of below 10⁻¹¹ when grown at 22 °C and at pH 5. The third and final containment system uses an excisionase based system to only express doc upon two, non-consecutive exposures to low pH, achieving a survival ratio of 10⁻⁴.

1.4. Evolutionary stability

Containment systems are only effective if they maintain functionality throughout the entire period of their intended use. Microbes exist across all domains of life, but are unified in their almost universal short generational lifespan and capacity to mutate and evolve at a rapid rate. This presents a problem for technology designed to contain and control the growth of engineered microbial species. To coin a term from the fictional Dr Ian Malcolm, “Life, uh, finds a way”. Any containment system that is designed to prevent microbial growth inherently has the potential to inflict a fitness defect, and therefore an evolutionary pressure to remove this fitness defect. For different forms of containment, this can occur in different manners.

For an auxotroph based containment system, an engineered strain would require the capacity to produce the metabolite that it is auxotrophic for, or negate the necessity for it in the first place, which can be achieved in four ways. 1) By taking on genes or operons from other strains that confer this capacity, through horizontal gene transfer or sexual reproduction, most likely from related species. 2) Evolution of pathways that are already intrinsic to the engineered strain that become capable of producing the required metabolite. 3) Evolution or modifications to the composition of the other organisms in the surrounding environment that increase production and/or excretion of the metabolites in question, allowing an environment that was initially non-permissible to become viable. 4) In the case of the systems mentioned that rely upon essential genes requiring ligand cofactors or nnAA for function/ translation (Lopez & Anderson, 2015; Mandell et al., 2015; Rovner et al., 2015), SNPs and small mutations to the genes in question can result in removing those requirements.

Escaping from a system of containment that uses the regulation of an essential gene can come about through three potential mechanisms. 1) Mutation of the regulation of the gene in question, making it no longer dependent upon whatever factor was designed to control expression. 2) Evolution of other pathways within the cell to fulfil the role the essential gene confers, negating the essential genes necessity. 3) Taking on the same or similar genes from other organisms that are regulated in an alternative manner.

Population control based upon the expression of a toxic gene has four possible modes of escape. 1) Mutation of the ORF of the toxin, negating its function 2) Mutation to the regulation of the toxin, either in its promoter or any transcriptional factors or in the regulation of the antitoxin (if present, see below) 3) Mutation in the target of the toxin. 4) Uptake of a resistance gene through horizontal gene transfer or sexual reproduction. Option 1, and in many cases option 2, represents a loss of function mutation, meaning the range of possible mutations that achieve the effect of escape is vastly wider than that required to confer a gain of function, which all other modes of escape mentioned above are some form of. Another factor that contributes to the instability of a containment system based on regulation of a toxic gene is the inherent leakiness of most regulatory systems. Promoters can best be described as up or down regulated rather than off or on, as even when repressed or not induced, low levels of expression can be observed. To counteract the fitness defect imparted from leaked expression of a toxic gene in permissible conditions, an antitoxin should be included in the design of the system. Most natural toxins evolved to function alongside their cognate antitoxin, and this dynamic can be exploited by expressing antitoxin at low levels to negate the effect of a leaked toxin (Gallagher et al., 2015; F. Stirling et al., 2017, 2020).

Of the 35 containment systems reported in Table 3, only 7 provide data of their long-term stability when passaged in permissible conditions. The multi-layered containment system (Gallagher et al., 2015), consisting of a biotin auxotrophy as well as the arabinose mediated regulation of the two essential genes ribA and glmS and the toxin ecoRI nuclease, was shown to be stable after passaging in permissible conditions for at least 110 generations. This system includes the ecoRI methyltransferase, the antitoxin to ecoRI nuclease. The Deadman and Passcode containment systems (Chan et al., 2015) also regulate the expression of ecoRI, as well as the essential gene murC, this time without the presence of ecoRI methyltransferase. Neither maintained their respective level of containment after a four-day period of growth, with an increase in survival ratio of 4–5 orders of magnitude. Passcode was additionally passaged for four days in E. coli MDS42pdu ΔrecA (Csörgo et al., 2012), a strain lacking recombinogenic and mobile genomic elements. This reduced the Passcode escapee rate by 3–5 logs over the four-day period. This practice is appropriate for certain purposes, although the fitness defect observed in E.coli MDS42pdu ΔrecA will prevent its widespread application.

The individual and combined essentializer, cryodeath and acidTRP containment systems (F. Stirling et al., 2017, 2020) all use a toxin antitoxin based approach, resulting in stable growth for over 100 generations. In addition, the construction of these systems explored a method for intelligent design of libraries with promoter variance to achieve the desired balance of expression between toxin and antitoxin.

1.5. Containment standard practices

Currently publications on containment systems are disparate in their reporting techniques. Not all reports display a quantifiable survival ratio by comparing cfu at permissible conditions to cfu at non-permissible conditions, and instead rely on metrics such as a growth curve to show growth disparities. Although it is easy to infer an effect from a growth curve, the fact that OD measurements do not differentiate well between living and dead cells means that a true survival ratio is hard to calculate. We recommend that wherever possible the survival ratios for containment systems should always be calculated by plating dilutions of culture in both permissible and non-permissible conditions and comparing cfu.

In most cases, individual studies report unique limits to the sensitivity of their survival assays. This limit predominantly comes from the total cfu that are given the possibility to grow in non-permissible conditions. Using the traditional techniques of spreading culture on agar, the upper limit of plating a bacterial culture is reached at about 10¹¹-10¹² cfu (Gallagher et al., 2015; Hirota et al., 2017; Mandell et al., 2015; Rovner et al., 2015; F. Stirling et al., 2020). Beyond this, the concentration of cells is so high as to cause clumping that prevents accurate assays. Increasing the area of agar plates used circumvents this, but quickly becomes prohibitive in the area of plates required. We recommend that when reporting the survival ratio for a containment system, it should become standard practice to reach this limit of detection. For some containment mechanisms, lower survival ratios could feasibly be detected using technology such as a morbidostat (Toprak et al., 2013). A morbidostat continually monitors the growth rate of a bacterial culture, automatically adjusting the concentration of a small molecule (often an antibiotic but theoretically any small molecule regulator) in response to culture density. By incrementally increasing/decreasing the concentration of a containment system regulator, an effective population of greater than 10¹⁶ can be exposed to a non-permissible condition. Although this capacity is currently not easily accessible for all labs, widespread adoption of this technique may become necessary as the field develops. It is difficult to claim a transgenic strain is effectively contained without improvements to the detection limit reached by plating techniques, unless such an environmental release involved a population several orders of magnitude less than the established survival ratio of the containment systems involved.

The majority of studies did not report data on the stability of their respective containment systems. For all containment systems, no matter the application, evolutionary stability is an essential quality, and can easily be assayed by passaging in permissible conditions while calculating the number of generations that pass. A comparison of survival ratios before and after this growth period allows for a simple display of evolutionary stability. We recommend that determining the stability of a strain over 100 generations should become standard practice when reporting a new containment system. This is sufficient to allow at least one adaptive sweep to pass through the population, allowing beneficial mutations to take over (Maddamsetti et al., 2015, Novick & Szilard, 1950). Additional experiments comparing the growth rates of engineered strains with parent strains, either individually or in competitive co-culture growth assays, would also be informative. Further theory on the capacity of asexual strains to maintain a circuit that confers only a very slight fitness disadvantage can be found in (Stirling et al., 2017).

Although NIH guidelines currently recommend a survival ratio of no greater than 10⁻⁸ for the containment of transgenic microbes, it is readily apparent that this standard is too high for many applications of transgenic microbes. A containment system should only be considered functional if its survival ratio is such that over the time period a transgenic microbe is to be deployed, the probability of an escape event occurring is negligible. For a transgenic population of microbes, if Y cells are expected to physically translocate away from their intended environment over the lifespan of the application, the survival ratio of a containment system must be orders of magnitude less than 10^−Y. Any application of transgenic microbes where greater than 10⁸ cells are expected to be deployed will require a survival ratio lower than the current guidelines for effective containment. Future recommendations could reflect the number of cells expected to survive, relating to both the population released and the escape rate of the containment systems involved.