Abstract

One problem with engineered genetic circuits and metabolic pathways in synthetic microbes is their stability over evolutionary time in the absence of selective pressure. We recently measured the evolutionary stability dynamics and determined the loss-of-function mutations for a wide variety of BioBrick-assembled genetic circuits in order to develop simple design principles for engineering mutationally robust genetic circuits (Sleight et al., 2010b). In this report, we focus on experiments performed with the LuxR receiver device, T9002, and reengineered versions of this circuit. T9002 loses function in fewer than 20 generations and the mutation that repeatedly causes its loss of function is a deletion between two homologous transcriptional terminators. To measure the effect between transcriptional terminator homology levels and evolutionary stability, we reengineered six versions of T9002 with a different transcriptional terminator at the end of the circuit. When there is no homology between terminators, the evolutionary half-life of this circuit is significantly improved by more than twofold and is independent of the expression level. Removing all homology between terminators and decreasing expression level fourfold increases the evolutionary half-life over 17-fold. We also investigated the predictability of loss-of-function mutations between nine replicate evolved populations. T9002 circuits reengineered to have no sequence similarity between terminators effectively removes a certain class of mutations from occurring, where the most common loss-of-function mutation is a deletion between 8-base pair (bp) scar sequences to remove the luxR promoter. Finally, we evolved T9002 and a reengineered version in media with different levels of inducer to measure the relationship between expression level and evolutionary stability. The results show that on average, like other circuits we studied, evolutionary half-life exponentially decreases with increasing expression levels.

Introduction

Experimental evolution is a powerful system for studying evolution in the laboratory over short or long timescales (Elena and Lenski, 2003). For these experiments, normally bacteria or other microbes are propagated over multiple generations in certain environmental conditions to understand the phenotypic and genetic differences between evolved strains and their progenitors (Herring et al., 2006; Riehle et al., 2005; Sleight and Lenski, 2007). Experimental evolution experiments also allow for the study of parallel genetic or phenotypic changes between replicate evolved populations. When multiple evolved populations converge on a similar phenotype, it is often a strong indicator of evolutionary adaptation, as opposed to the product of genetic drift (Bull et al., 1997; Colosimo et al., 2005; Sleight et al., 2008). Understanding convergent evolution is of interest because it increases our ability to predict the evolution of unknown strains in different environments.

Synthetic biologists use engineering principles to design and construct new organisms that do not exist in nature (Endy, 2005). Often these synthetic organisms are programmed with genetic circuits that are constructed from biological parts such as promoters, ribosome binding sites, coding sequences, and transcriptional terminators called BioBricks (Endy, 2005; Shetty et al., 2008). The MIT Registry of Standard Biological Parts (called “the Registry” from here on) maintains over 3,000 BioBricks encoded on plasmids that are available to researchers with a wide variety of functions, from bacterial photography, to quorum sensing, to odor production and sensing. BioBricks are widely available for the design of more complex systems but in general are not well characterized (Canton et al., 2008; Endy, 2005). The most well characterized part to date is a cell-to-cell communication receiver device (Canton et al., 2008), which was provided with a published prototype “biological part data sheet” containing information engineers would need to use it in their own designs. One of the figures in this data sheet describes the reliability of this circuit over evolutionary time. Connecting the receiver device to a green fluorescent protein (GFP)-reporter device causes this circuit to repeatedly lose function in fewer than 100 generations due to a deletion mutation between transcriptional terminators that are repeated in both the receiver and reporter devices.

From a synthetic biologist’s perspective, ideally synthetic organisms will be robust to environmental conditions and mutations to allow for optimal function over evolutionary time. However, the genetic circuits encoded on plasmids in synthetic organisms are often unstable over evolutionary time if there is no selective pressure to maintain function of the circuit. This loss of function occurs because a cell in the population that has a mutation in one of its plasmids may have a slight growth advantage. When this cell divides, the daughter cell may acquire more copies of the mutant plasmid, giving it a greater growth advantage. Over multiple generations, the functional cells in the population may be outcompeted by nonfunctional cells, unless these evolutionary events are controlled.

From the public health perspective, if a synthetic microbe escaped the laboratory into the environment, how long would it survive and how many generations would it take before its synthetic function was lost over evolutionary time? Could a synthetic pathogenic strain evolve to become more virulent over time? To our knowledge, these questions are largely unexplored, but from this perspective we hope that synthetic cells will not be robust to environmental conditions and mutations. As a first step toward understanding the evolutionary stability dynamics and predictability of loss-of-function mutations in genetic circuits, we evolved replicate populations of Escherichia coli carrying several different genetic circuits for multiple generations (Sleight et al., 2010b).

In this study, we found that a wide variety of loss-of-function mutations are observed in BioBrick-assembled genetic circuits including point mutations, small insertions and deletions, large deletions, and insertion sequence (IS) element insertions that often occur in the scar sequence between parts. Promoter mutations are selected more than any other biological part. We also found that, on average, evolutionary half-life exponentially decreases with increasing expression levels. Surprisingly, one particular circuit that was expressed at a low level due to having a weak promoter was able to maintain function for 500 generations, but most circuits lost function within 100 generations. We were able to reengineer genetic circuits to be more mutationally robust using a few simple design principles: high expression of genetic circuits comes with the cost of low evolutionary stability, avoid repeated sequences, and the use of inducible promoters increases stability. Interestingly, inclusion of an antibiotic resistance gene within the circuit does not ensure evolutionary stability, but a reengineered version of this circuit may improve stability when certain environmental conditions are used.

In this report, we focus on the results of this study from experiments performed with the LuxR receiver device, T9002, and six reengineered versions of this circuit. We then discuss the relevance of these results for emerging microbial pathogens and future directions.

Results

Loss-of-Function Mutations and Evolutionary Stability Dynamics in the T9002 Circuit

We first measured the evolutionary stability dynamics of the T9002 genetic circuit (Canton et al., 2008) propagated in Escherichia coli MG1655 in order to determine the loss-of-function mutations that cause its instability. High-copy plasmids were used instead of low- or medium-copy plasmids to maximize selective pressure so that evolution would occur more rapidly since replication and expression of genetic circuits encoded on high-copy plasmids will increase metabolic load and lower fitness. Cells with a low metabolic load (e.g., cells with mutant plasmids) have greater fitness than cells with a higher metabolic load (e.g., cells with functional plasmids) (unpublished results). Therefore, we expect that mutants will be able to rapidly outcompete functional cells that have a high expression level. However, other factors besides expression level will play a role in this evolutionary process such as mutation rate and the metabolic load associated with plasmid replication.

T9002 is a Lux receiver circuit that expresses luxR that activates GFP expression when the inducer AHL is added to the media (Figure A18-1). The evolutionary stability dynamics were measured by serial propagation with a dilution factor that allows for about 10 generations per day. Figure A18-2 shows the evolutionary stability dynamics of the T9002 circuit propagated in high-input (with AHL) and low-input (without AHL) conditions. From different time points in the experiment, the low- and high-input populations were induced with AHL to measure their normalized expression (here measured by fluorescence divided by cell density) over time. The low-input evolved populations slowly lose their function to about 50 percent of the maximum after 300 generations. The evolved populations in high-input conditions rapidly lose their function in fewer than 30 generations. The dynamics of this evolutionary stability are described below in Figure A18-5. No functional clones were observed after 30 generations as determined by measurement of individual colonies. The mutation that repeatedly causes loss of function in the high-input evolved populations is a deletion between two homologous transcriptional terminators (Figure A18-3), the same mutation described by Canton et al. (2008). This mutation evidently occurs at such a high rate that mutants quickly take over the population. In fact, Canton et al. (2008) were unable to isolate a population derived from a single isolate that did not already carry the deletion. The mutant plasmid was transformed back into the progenitor and was shown not to fluoresce after induction with AHL.

FIGURE A18-1

The T9002 genetic circuit. Symbols depict promoters (bent arrows), ribosome binding sites (ovals), coding sequences (arrows), and transcriptional terminators (octagons). T9002 consists of two devices, a LuxR receiver device and a GFP-expressing device. (more...)

FIGURE A18-2

Evolutionary stability dynamics of T9002 evolved under low-input (−AHL) and high-input (+AHL) conditions. Low- and high-input evolved populations are shown with solid gray circles and solid black circles, respectively. Evolved populations at different (more...)

FIGURE A18-5

Evolutionary stability dynamics of T9002 and reengineered T9002 circuits. The fluorescence/OD₆₀₀ is plotted versus the generations for T9002 (solid black circles) and T9002 reengineered circuits (various shapes and colors) under high-input (+AHL) conditions. (more...)

FIGURE A18-3

T9002 loss-of-function mutation. This circuit repeatedly has a deletion between homologous repeated terminators after 30 generations in the high-input evolved populations. This is the same mutation found by Canton et al. (2008). SOURCE: This figure is (more...)

Evolution Experiment with Reengineered Circuits

Based on the results of the previous experiments, we reengineered the T9002 circuit to test various predictions of evolutionary stability and mutational predictability. The loss-of-function mutation was a deletion that repeatedly occurred between two homologous transcriptional terminators. Mutations and genetic rearrangements can occur due to misalignment of homologous sequences during replication (termed “replication slippage”) (Lovett, 2004). Deletion mutations between repeated sequences are known to be dependent on repeat length, proximity, and homology level (Bzymek and Lovett, 2001). These deletions are recA-independent if the repeat length is less than 200 bp (Bi and Liu, 1994; Lovett, 2004), as is the case with the repeated terminators in the T9002 circuit. Thus, we reengineered the last terminator of T9002 with various terminators available in the Registry to measure the effect of terminator homology level and orientation with evolutionary stability (Figure A18-4). We predicted that we could increase evolutionary robustness by decreasing the mutation rate of this deletion. Furthermore, although there have been several studies on recombination between repeated sequences, this phenomenon has not been studied in the context of synthetic biology using genetic circuits constructed from BioBricks. For instance, we do not know the effect of using various BioBrick terminators with different homology levels in the same circuit. The use of different terminators will become increasingly important when more complex circuits are constructed and Bio-Bricks become even more widespread in the community. Also, many of the studies on recombination between repeated sequences use antibiotic resistance genes to measure recombination rates and may not relate to actual functioning circuits.

FIGURE A18-4

T9002 reengineering. T9002 reengineering involves changing the second double transcriptional terminator with varying degrees of homology and orientation to the first double transcriptional terminator. SOURCE: This figure is taken from the Journal of Biological (more...)

Since we learned from previous experiments that evolutionary stability dynamics of genetic circuits have high variability between replicate populations, we evolved nine independent populations of each reengineered circuit for at least 300 generations. Three experimental replicate populations of three independent transformants were used to test for independent mutational events. A single transformant may have a mutation at a low level that will eventually sweep through the population, so if only one transformant was used, the same mutation may show up in all replication populations. For each of the nine populations in every circuit, the evolutionary half-life was measured to quantitate the number of generations until the expression level had decreased to half of its initial level (Table A18-1). Plasmids from a single clone from each evolved population were then sequenced after the population level had decreased to less than 10 percent of the original expression level, or after 500 generations, whichever came first.

TABLE A18-1

Evolutionary Half-Life of T9002 and Reengineered T9002 Genetic Circuits.

Reengineered T9002 Circuits with Different Transcriptional Terminators: Loss-of-Function Mutations and Evolutionary Stability Dynamics

Figure A18-4 shows the schematic for reengineering the last transcriptional terminator in the T9002 circuit. The evolutionary stability dynamics for six reengineered T9002 circuits and the original T9002 circuit are shown in Figure A18-5. Figure A18-6 shows the types of mutations that occurred in each of the nine replicate evolved populations. Finally, Figure A18-7 shows the most common mutations for each reengineered circuit. The six reengineered circuits are labeled T9002-A through T9002-F in Figure A18-5 and are color coded to correspond to the same circuit mutations shown in Figure A18-7. These circuits were all propagated with the inducer AHL to allow for GFP expression.

FIGURE A18-6

Loss of mutations in nine independently evolved populations. For nine independently evolved populations, colored boxes correspond to the mutation legend below the table. The most common mutation for a particular type of mutation is labeled with “1” (more...)

FIGURE A18-7

Most common loss-of-function mutations in reengineered T9002 circuits. The original circuit is shown above the loss-of-function mutation circuit for each of the six reengineered T9002 circuits. SOURCE: This figure is taken from the Journal of Biological (more...)

In the following paragraphs, the loss-of-function mutations and evolutionary stability dynamics for the original T9002 circuit and for each reengineered circuit are described in detail.

T9002 The original T9002 circuit decreases rapidly to about 30 percent of the original level after only 10 generations and all function is lost by 20 generations (Figure A18-5). The same deletion between homologous terminators as was observed in previous experiments (Figure A18-3) was found in all nine replicate populations (Figure A18-6). The evolutionary half-life of this circuit was found to be about 7.1 generations on average (Table A18-1).

T9002-A The final double terminator in T9002 was reengineered in the reverse complementary orientation (called B0025 in the Registry) to make T9002-A. The stability of this circuit has approximately the same expression level and stability dynamics as T9002, but it has an evolutionary half-life of about 5.6 generations (Figure A18-5, Table A18-1). This decreased stability may be because the terminator in the reverse orientation is more likely to cause replication slippage. Since the expression level is similar to T9002 and therefore the metabolic load should be roughly equivalent, the difference in stability is primarily due to an increased mutation rate. Seven of nine replication populations have a deletion between the first B0010 terminator and the reverse complement of B0010 (Figures A18-6 and A18-7). This effect likely occurs because B0010 has a long hairpin structure, so one-half of B0010 can interact with the other half of the reverse complementary B0010 sequence during DNA replication since they have perfect homology. Two of the nine populations had a deletion that formed a triple terminator of B0010-B0012-B0010.

T9002-B The T9002-B circuit was reengineered to rearrange the B0010 and B0012 terminators to have B0012 first and then B0010. The rearrangement decreases the expression level to almost zero initially and this expression drifts up over time and then decreases to zero (Figure A18-5). For this circuit, evolutionary half-life measurements are essentially meaningless due to the randomness of low expression. Notice that other reengineered T9002 circuits also have decreased expression levels relative to T9002, presumably due to weaker terminator hairpin structures having increased mRNA degradation (Smolke and Keasling, 2002) or transcriptional readthrough that can decrease plasmid copy number (Stueber and Bujard, 1982). Others have observed that removal of transcriptional terminators can decrease expression levels in general (Kaur et al., 2007). All nine populations have the same deletion between B0010 terminators (Figures A18-6 and A18-7). Because the B0010 terminator is an inexact hairpin (there are some mismatches), one-half of the first B0010 interacts with the other half of the second B0010 terminator, causing a hybrid B0010 terminator (Figure A18-7).

T9002-C The T9002-C circuit was reengineered to have B0012 and B0011 as the final double terminator instead of B0010 and B0012. This circuit has nearly identical stability dynamics as T9002, with an evolutionary half-life of about 6.7 generations on average (Figure A18-5, Table A18-1). All nine populations have the same deletion between B0012 terminators that make a triple terminator of B0010-B0012-B0011 (Figures A18-6 and A18-7). Since the expression level and stability dynamics are roughly equivalent to that of T9002, the mutation rate between repeated B0010-B0012 terminators (129 bp) is probably about the same as that between repeated B0012 terminators (41 bp). Interestingly, no significant stability difference was observed between T9002-C (41-bp homology) and T9002 or T9002-A (both 129-bp homology), despite having similar expression levels. This result suggests that shortening the repeated regions of homologous terminators did not increase evolutionary robustness, contrary to what we expected.

T9002-D The T9002-D circuit has the same final B0012-B0011 terminator but is the reverse complement of this sequence. The inclusion of this terminator decreases the initial expression level to about 65 percent of T9002 (Figure A18-5). The evolutionary half-life of this circuit is about 57 generations (Figure A18-5, Table A18-1). Also, the slope of the stability plot is decreased relative to other circuits with higher expression (T9002, T9002-A, T9002-C, and T9002-E) and the stability lag (time for expression to decrease to zero along the x axis) is increased (Figure A18-5). In contrast to other circuits with repeated terminators, only three of nine have deletions between homologous terminators, forming a hybrid B0012 terminator (Figures A18-6 and A18-7). This result is probably because, unlike T9002-C, the second B0012 is the reverse complement, and therefore the only homology in this circuit is the 8-bp hairpin structure having complementary sequences; the rest of the terminator has a loop structure of noncomplementary sequences. In other words, B0010 has sufficient homology in either the forward or reverse orientation to cause replication slippage, but in B0012 replication slippage is more likely to occur only in the forward orientation. The other mutations in this circuit are composed of point mutations, short insertions or deletions, deletions between scar sequences, or IS mutations (Figure A18-6).

T9002-E The T9002-E circuit, like the T9002-F circuit, was reengineered to have no homology between terminator sequences. This circuit has the highest initial expression level on average probably because J61048 is a very strong terminator, but has similar expression relative to the T9002, T9002-A, and T9002-C circuits (Figure A18-5). Its evolutionary half-life is about 16.7 generations (Figure A18-5, Table A18-1). Thus, relative to other circuits with similar expression levels, it is the most evolutionary robust circuit, having over twofold higher stability than T9002. When the evolutionary half-life is measured for the nine replicate populations, this evolutionary half-life difference compared to T9002 is highly significant (one-tailed t-test, p = 0.0003). Notice that the stability slope is similar to that for T9002, T9002-A, and T9002-C circuits, but the stability lag is increased by about 10 generations. This difference in lag is likely due to a decreased mutation rate since mutations are more random compared to the other similar expression-level circuits (Figure A18-6). The most common mutation is a deletion between repeated scar sequences that removes the luxR promoter and effectively inactivates the circuit function (Figure A18-7).

T9002-F The T9002-F circuit was reengineered with the B0011 terminator, so it also has no homology between terminator sequences. The B0011 is evidently a weak terminator since its initial expression level is about fourfold lower than that of T9002. Its stability dynamics show that it is the most evolutionary robust of the reengineered T9002 circuits, with an evolutionary half-life of about 125 generations (Figure A18-5, Table A18-1). This result indicates that decreasing homology levels and expression through terminator reengineering increased the evolutionary half-life of this circuit over 17-fold relative to T9002. Like T9002-E, the mutations in each of the nine populations are mostly random (Figure A18-6). Also like T9002-E, the most common mutation is a deletion between repeated scar sequences that removes the luxR promoter driving GFP expression (Figure A18-7). Since T9002-E and T9002-F likely have similar mutation rates with zero terminator homology, the large stability difference between these circuits can be explained by expression levels alone.

Overall, excluding T9002-B, three of the five reengineered T9002 circuits are more evolutionary robust than the original circuit. The order of evolutionary robust genetic circuits is as follows: T9002-F > D > E > A = C = T9002. This increase in evolutionary robustness can be attributed to decreased expression levels (due to the terminator reengineering) and to decreased mutation rate between homologous transcriptional terminators. The reengineered circuits with homologous transcriptional terminators almost always have deletions between homologous regions, whereas circuits without homology have mutations in other locations in the circuit. Reengineering this circuit to remove all homology effectively removes a certain class of mutations from occurring. The T9002-E circuit is more evolutionarily robust than other circuits with similar expression levels, which is likely due to decreased mutation rate alone. Thus, evolutionary robustness can be increased by removing long repeated sequences from genetic circuits, but even short 8-bp scar sequences have the potential for replication slippage.

Evolutionary Half-Life Versus Initial Expression Level in T9002 and T9002-E Circuits Evolved with Different Inducer Concentrations

From previous results (Figure A18-5, Table A18-1), we noticed that circuits with a high initial expression level tended to have low evolutionary stability. Also, particular circuits with high mutation rates had lower stability compared to circuits with lower mutation rates. To test the relationship between initial expression level and evolutionary half-life directly, we evolved the T9002 (high mutation rate) and T9002-E (lower mutation rate) circuits propagated with different AHL concentrations.

Figure A18-8 shows the mean initial expression level versus mean evolutionary half-life for eight replicate populations from three different AHL concentrations in T9002 (black) and T9002-E (red). An exponential fit of these mean data points gives a much higher r² value than a linear fit (>0.1) in both cases. T9002 has an r² value of 0.954 compared to the r² value of 0.955 in T9002-E. The curve for T9002 is shifted to the left from T9002-E due to its higher mutation rate (expression alone cannot account for the shift), but as expression is decreased the evolutionary half-life difference between these two circuits also decreases. This decrease may be because, at high expression levels, the fitness difference between the progenitor and mutant cells is the highest, and therefore mutants outcompete functional cells in the population more quickly.

FIGURE A18-8

Evolutionary half-life versus initial expression level in T9002 and T9002-E circuits evolved with different inducer concentrations. Evolutionary half-life versus initial expression level is plotted for T9002 (solid black circles) and T9002-E (solid dark (more...)

Discussion

Genetic circuits lose function over evolutionary time and are found to have a wide variety of mutations that cause their loss of function. Circuits with repeated sequences almost always have deletions between these sequences, but the effect of repeat length is not well understood. In one reengineered T9002 circuit, shortening the length of homology from 129 to 41 bp did not significantly increase evolutionary stability. Stability was only increased when there was no homology whatsoever between transcriptional terminators. Mutations between repeated sequences without perfect homology in the case of some reengineered T9002 circuits are usually, but not always, predictable.

In circuits without repeated sequences, mutations are more random between evolved replicate populations. In other circuits we studied (not shown here), mutations that remove promoter function are most often selected for among all the genetic circuits tested (Sleight et al., 2010b). This result is likely because promoter mutations remove the metabolic load at both the transcriptional and translational levels. Mutations within RBSs are not found and mutations in coding sequences are rare except when that coding sequence is an activator of transcription downstream (as in the case of the luxR coding sequence in T9002). In the case of T9002, removing homology between transcriptional terminators only shifts the mutation to one that often removes function of the luxR promoter or luxR coding sequence instead.

What is needed is a method to predict the evolutionary stability of circuits from the properties of their parts, but the emergent behaviors of circuits will likely make prediction difficult. At the very least, publishing the evolutionary stability properties of simple circuits in future data sheets may allow engineers to calculate the expected evolutionary stability of more complex circuits. This calculation would likely require software (Chandran et al., 2009) and mathematical modeling (Chandran et al., 2008) that analyzes each part individually and the entire DNA sequence as a whole to determine the expected evolutionary stability. This calculation would also require standardization for methods to measure evolutionary stability, and the methods described here are not necessarily the best way. On the other hand, more general methods may be developed that focus less on design of the circuit and more on design of the environment to impose a selective pressure for function of the circuit (Bayer et al., 2009). Design of a selective environment is ideal, but it is difficult to do when the output of most circuits (e.g., GFP) is not linked to survival or growth rate. A cell-sorter device that sorts between functional and nonfunctional cells may help with this issue, but it may not be practical for performing routine experiments.

From our results of what types of mutations are selected for in genetic circuits and the evolutionary stability dynamics, a few simple design principles can be proposed when engineering circuits. The first principle is that high expression of genetic circuits comes with the cost of low evolutionary stability. Although exceptions to this rule certainly occur, a genetic circuit with high expression correlates with a large metabolic load and therefore is predicted to have decreased cellular fitness. When the fitness difference between the functional and nonfunctional cells in the population is large, evolutionary stability will decrease quickly. Therefore, the initial expression level of the circuit is likely to be a good predictor of evolutionary stability when a circuit with high mutational robustness is desired. Using a low- or medium-copy plasmid will help with stability as long as the expression level does not need to be high. For more complex circuits where a high expression level is needed for proper functioning of the circuit, decreasing expression level then comes at the cost of changing the function of the circuit.

The second design principle is to avoid repeated sequences. This principle may be obvious, but nearly every circuit in the Registry with more than one coding sequence has repeated B0015 terminators. When a circuit has a high metabolic load (higher than T9002) and repeated sequences on a high-copy-number plasmid, when using MG1655 as the host strain, the circuit will almost always lose function during overnight growth (unpublished results). Reengineering the T9002 circuit to have two different transcriptional terminators (T9002-E) does significantly increase evolutionary half-life by more than twofold and is independent of expression levels. However, since this circuit has high expression, this improvement only results in an increase of about 10 generations. Decreasing the expression level along with the mutation rate will increase the evolutionary half-life about 17-fold, as seen in the T9002-F circuit. This result suggests that simple ways to increase evolutionary stability can be used without changing the function of the circuit. For more complex circuits, the community will need to identify many more terminators than those that currently exist in the Registry to design circuits without repeated sequences.

The third design principle is that use of inducible promoters generally increases evolutionary stability (not shown here) (Sleight et al., 2010b). This principle may or may not be significant depending on the circuit used. Inducible circuits are likely more stable due to decreased metabolic load and are preferred since expression can be controlled and fine-tuned, though in some circumstances a constitutive promoter may be desired. Therefore, the use of inducible promoters can be thought of as one extra precaution to maximize evolutionary stability, but expression levels and repeated sequences should first be considered.

We emphasize that the design principles proposed may not be general since only relatively simple circuits were tested in this study. Evolutionary stability should be measured in larger and more complex circuits to understand if these design principles apply. Furthermore, these simple design principles should not necessarily be all used simultaneously. A researcher may not want only to design circuits that have low expression, have no repeated regions, and use a promoter that is inducible. For instance, if recombination sites are needed in the circuit, then repeated or inverted sequences may be impossible to avoid. In addition to the design for the proper function of the circuit, the design for evolutionary robustness should be carefully considered. For this, we need to think about the probability of mutations occurring when the expression level, and therefore metabolic load, is high. In this study, removing repeated regions often shifts mutations to the promoter, and putting a selection on the promoter often shifts the mutation to the chromosome (not shown here) (Sleight et al., 2010b).

Thus, mutations are unavoidable without a selective pressure, but evolutionary stability can likely be improved in the future by better design of selective environments where the circuit is linked to survival and/or growth rate, understanding of mutation rates in genetic circuits, fitness differences between functional and nonfunctional cells, and improvements to the host strain that decrease mutation rates or buffer metabolic loads more efficiently. Another way to improve evolutionary stability is to engineer an error detection and correction circuit that will correct mutations, but it will need careful design since this circuit itself will be prone to mutation. Recently, a technique called ClChE was developed to insert and amplify (increase copy number of) genetic circuits on the chromosome to avoid the problem of maintaining evolutionary stability of function on plasmids (Tyo et al., 2009). This is a powerful method that should be used, but it is also somewhat tedious to insert and amplify circuits on the chromosome. There are also examples of chromosomal genes that are evolutionarily unstable when not under selection (Cooper and Lenski, 2000; Cooper et al., 2001) and therefore chromosomally integrating synthetic circuits will only delay this problem. Designing mutationally robust genetic circuits, therefore, is somewhat of an art form at the moment besides a few simple design rules, but it should be seen as something the engineer can eventually control.

From the opposite perspective, if a synthetic microbe is released into the environment, we need to hope it will not be mutationally robust. For this, we need to be able to predict how it will evolve if it is able to survive and compete with microbes in their natural environment. One circuit with a weak constitutive promoter was stable for 500 generations in the presence of ampicillin (Sleight et al., 2010b), so it is theoretically possible for circuits to maintain their function for many generations in the right environmental conditions. For the T9002 circuits described here, if the cells carrying these plasmids escaped into the environment, there would likely not be any ampicillin to maintain the plasmid itself, so plasmid stability may follow similar dynamics, as shown by Lenski and Bouma (1987). As far as circuit stability goes, since it is unlikely there will be AHL in the environment to keep circuit function on, function of the circuit will likely decay with similar dynamics as that shown in Figure A18-2 in the T9002 populations evolved without AHL. A circuit with a lower expression level and lower mutation rate (such as T9002-F) will decay more slowly, but it really depends on the interaction between circuit function and the selective pressure in a particular environment.

When thinking about the evolutionary trajectory of a synthetic microbe, if the circuit function imparts a selective advantage over natural microbes, it should be of concern. For example, it is unlikely that GFP will provide a fitness boost in a particular environment, but a different circuit that allows for the degradation of a carbon source may allow cells with this function to have a growth advantage and permanently change an ecosystem. More worrisome is the release of a synthetic microbe that evolves to become pathogenic by acquiring genes from a virus, for instance. Here, one could imagine cells with T9002 mutating out the GFP part of the circuit and inserting a virulence factor instead. Then, the new cells would express the virulence factor only if AHL was present—a remote possibility if there are other microbes that use this molecule for quorum sensing, but still possible. The results shown in this report illustrate the evolution of genetic circuits when no selective pressure is imposed to maintain function of the circuit, but it is unknown what will happen in a natural environment where real selective pressures exist. The study of natural pathogen evolution in E. coli and other pathogens in combination with evolutionary studies in synthetic biology will allow us to understand the risks involved.

Acknowledgments

We thank members of the Sauro and Klavins labs for materials and useful discussions. A special thanks goes to Lucian Smith, Wilbert Copeland, and Kyung Kim for valuable suggestions. We also thank three anonymous reviewers for valuable feedback. This research was funded by the National Science Foundation (NSF) Grant in Theoretical Biology (0827592), and BEACON: An NSF Center for the Study of Evolution in Action (0939454). Part of this research was previously published in the Journal of Biological Engineering. This journal is an open-access journal where the authors retain copyright and hold an Open Access license (http://www.biomedcentral.com/info/authors/license).

Methods

References

• Bayer TS, Hoff KG, Beisel CL, Lee JJ, Smolke CD. Synthetic control of a fitness tradeoff in yeast nitrogen metabolism. Journal of Biological Enginerring. 2009;3:1. [PMC free article: PMC2631470] [PubMed: 19118500]
• Bi X, Liu LF. recA-independent and recA-dependent intramolecular plasmid recombination. Differential homology requirement and distance effect. Journal of Molecular Biology. 1994;235(2):414–423. [PubMed: 8289271]
• Bull JJ, Badgett MR, Wichman HA, Huelsenbeck JP, Hillis DM, Gulati A, Ho C, Molineux IJ. Exceptional convergent evolution in a virus. Genetics. 1997;147(4):1497–1507. [PMC free article: PMC1208326] [PubMed: 9409816]
• Bzymek M, Lovett ST. Instability of repetitive DNA sequences: The role of replication in multiple mechanisms. Proceedings of the National Academy of Sciences of the United States of America. 2001;98(15):8319–8325. [PMC free article: PMC37438] [PubMed: 11459970]
• Canton B, Labno A, Endy D. Refinement and standardization of synthetic biological parts and devices. Nature Biotechnology. 2008;26(7):787–793. [PubMed: 18612302]
• Chandran D, Copeland WB, Sleight SC, Sauro HM. Mathematical modeling and synthetic biology. Drug Discovery Today: Disease Models. 2008;5(4):299–309. [PMC free article: PMC5102263] [PubMed: 27840651]
• Chandran D, Bergmann FT, Sauro HM. TinkerCell: Modular CAD tool for synthetic biology. Journal of Biological Engineering. 2009;3:19. [PMC free article: PMC2776589] [PubMed: 19874625]
• Colosimo PF, Hosemann KE, Balabhadra S, Villarreal G, Dickson M, Grimwood J, Schmutz J, Myers RM, Schluter D, Kingsley DM. Widespread parallel evolution in stickle-backs by repeated fixation of Ectodysplasin alleles. Science. 2005;307(5717):1928–1933. [PubMed: 15790847]
• Cooper VS, Lenski RE. The population genetics of ecological specialization in evolving Escherichia coli populations. Nature. 2000;407(6805):736–739. [PubMed: 11048718]
• Cooper VS, Schneider D, Blot M, Lenski RE. Mechanisms causing rapid and parallel losses of ribose catabolism in evolving populations of Escherichia coli B. Journal of Bacteriology. 2001;183(9):2834–2841. [PMC free article: PMC99500] [PubMed: 11292803]
• Elena SF, Lenski RE. Evolution experiments with microorganisms: The dynamics and genetic bases of adaptation. Nature Reviews Genetics. 2003;4(6):457–469. [PubMed: 12776215]
• Endy D. Foundations for engineering biology. Nature. 2005;438(7067):449–453. [PubMed: 16306983]
• Herring CD, Raghunathan A, Honisch C, Patel T, Applebee MK, Joyce AR, Albert TJ, Blattner FR, van den Boom D, Cantor CR, Palsson B. Comparative genome sequencing of Escherichia coli allows observation of bacterial evolution on a laboratory timescale. Nature Genetics. 2006;38(12):1406–1412. [PubMed: 17086184]
• Kaur J, Rajamohan G, Dikshit KL. Cloning and characterization of promoter-active DNA sequences from Streptococcus equisimilis. Current Microbiology. 2007;54(1):48–53. [PubMed: 17171467]
• Lenski RE, Bouma JE. Effects of segregation and selection on instability of plasmid pACYC184 in Escherichia coli B. Journal of Bacteriology. 1987;169(11):5314–5316. [PMC free article: PMC213944] [PubMed: 3312174]
• Lovett ST. Encoded errors: Mutations and rearrangements mediated by misalignment at repetitive DNA sequences. Molecular Microbiology. 2004;52(5):1243–1253. [PubMed: 15165229]
• Riehle M, Bennett AF, Long AD. Changes in gene expression following high-temperature adaptation in experimentally evolved populations of E. coli. Physiological and Biochemical Zoology. 2005;78(3):299–315. [PubMed: 15887077]
• Shetty RP, Endy D, Knight TF. Engineering BioBrick vectors from BioBrick parts. Journal of Biological Engineering. 2008;2:5. [PMC free article: PMC2373286] [PubMed: 18410688]
• Sleight SC, Lenski RE. Evolutionary adaptation to freeze-thaw-growth cycles in Escherichia coli. Physiological and Biochemical Zoology PBZ. 2007;80(4):370–385. [PubMed: 17508333]
• Sleight SC, Orlic C, Schneider D, Lenski RE. Genetic basis of evolutionary adaptation by Escherichia coli to stressful cycles of freezing, thawing and growth. Genetics. 2008;180(1):431–443. [PMC free article: PMC2535694] [PubMed: 18757947]
• Sleight SC, Bartley BA, Lieviant JA, Sauro HM. In-Fusion BioBrick assembly and re-engineering. Nucleic Acids Research. 2010;38(8):2624–2636. [PMC free article: PMC2860134] [PubMed: 20385581]
• Sleight SC, Bartley BA, Lieviant JA, Sauro HM. Designing and engineering evolutionary robust genetic circuits. Journal of Biological Engineering. 2010;4(1):12. [PMC free article: PMC2991278] [PubMed: 21040586]
• Smolke CD, Keasling JD. Effect of gene location, mRNA secondary structures, and RNase sites on expression of two genes in an engineered operon. Biotechnology and Bioenginerring. 2002;80(7):762–776. [PubMed: 12402322]
• Stueber D, Bujard H. Transcription from efficient promoters can interfere with plasmid replication and diminish expression of plasmid specified genes. EMBO Journal. 1982;1(11):1399–1404. [PMC free article: PMC553223] [PubMed: 6327267]
• Tyo KEJ, Ajikumar PK, Stephanopoulos G. Stabilized gene duplication enables long-term selection-free heterologous pathway expression. Nature Biotechnology. 2009;27(8):760–765. [PubMed: 19633654]

87

Corresponding author: Sean C. Sleight, e-mail: ude.notgnihsaw.u@thgiels.

88

Department of Bioengineering, University of Washington, Seattle, WA 98195, USA.