A major motivation for adopting minimal genome chassis organisms in applications of synthetic biology centers on minimising mutations and genetic diversity and thus enhancing stability and reproducibility. These are desirable properties in biotechnologies and there is promising lab-scale evidence [6, 9, 15, 16] that minimal chassis can deliver them for white-biotechnologies, where the environment can be tightly controlled [12, 13]. However, for many proposed applications of synthetic biology, such as in environmental sensors, organisms will be exposed to environmental stress, such as metabolic stress. Evidence from both lab and natural bacterial populations show that stresses increase mutation rates [21,22,23]. An increase in stress-induced mutagenesis in an engineered organism would have potentially detrimental effects on the organism’s stability and persistence.
Our results show that even in as little as 24 h of metabolic stress, the MDS42 strain accumulated mutations just as quickly as MG1655, a strain that was not engineered for genetic stability. Despite marked variation within the runs, the estimated mutation rates rose in both strains suggesting that minimising a genome does not offer the benefits of stability during prolonged metabolic stress. Moreover, the prevalence of slow growth rates for mutants from both strains at all time points were indicative of a drop in fitness. The average biomass yield for both MDS42 and MG1655 mutants were either similar to or lower than yields for non-mutants (Additional file 1: Figure S3) suggesting mutants were unlikely to be adaptive. For the MG1655 mutants we also showed that whilst mixed cultures of mutants grew faster than monocultures their cumulative growth rate did not exceed rates observed in non-mutants. Therefore, taken together the data suggest that the accumulation of mutants is more likely attributable to a continued supply of slow-growing new mutants rather than a few fast growers. Slow growing mutants are more likely to be washed out of the chemostat and therefore our samples are likely to underestimate mutation accumulation. So, the true rate of mutation in our stressed populations is expected to be even higher. Indeed, to demonstrate just how conservative our estimate of mutation rate is we considered the probability that the average mutant is washed out of the chemostat. Suppose that mutants randomly appear and then grow at a rate η
m
. Letting N(t) be the number of individuals of a particular mutant, t hours after they first appeared. Then assuming N(t) is continuous,
$$\frac{dN(t)}{dt} = \eta_{m} N(t) - \frac{Q}{V}N(t),$$
(1)
where Q is the flowrate into and out of the chemostat and V is the volume and hence, Q/V is the dilution rate, which fixes the mean growth rate. So, replacing Q/V with η the growth rate of the non-mutant majority and solving we get
$$N(t) = N_{0} \text{e}^{{(\eta_{m} - \eta )t}} ,$$
(2)
where N0 = N(t = 0). Given that in our case N0 = 1 if we assume that births and loss occur randomly in the period t we can approximate the probability, P, that the mutant population has left by P = 1 − N(t) and so,
$$1 - P = \text{e}^{{(\eta_{m} - \eta )t}} .$$
(3)
Therefore, we can estimate the time taken for the probability of washout of the mutant subpopulation to be P as
$$t = \frac{\ln (1 - P)}{{(\eta_{m} - \eta )}}$$
(4)
In our case the dilution rate and hence growth rate (η) is 0.1 h−1. We assume that the mutant, with lower growth rate, does not become abundant enough to affect the overall population growth rate and hence η remains constant. The average mutant in the MDS42 strain grows half as fast as a non-mutant in the chemostat and therefore, η
m
= 0.05 h−1. So with probability, P = 0.99, the mutants will have washed out in t = 92.1 h. Thus, when mutants grow half as quickly as the general population, then we can be 99% confident that those that we see in a sample appeared in the population within the past 92 h (Additional file 1: Figure S4 for the full distribution). So in our case when we see 6.52 × 107 mutants (the average of the three MDS42 chemostat runs at 21 days) in the population, a mutant has appeared on average every 0.005 s.
It is generally thought that new mutations are usually neutral or deleterious and beneficial mutations are very rare ([36]). Moreover if the mutation has occurred in a gene that only makes a small contribution to a particular phenotype, the mutation might appear ‘silent’, with little or no discernible effect over the (non-mutation driven) major transcriptional changes that make large contributions to the overall phenotype under investigation. Indeed, metabolic stress and fluctuating reactor conditions elicit transcriptional changes in chemostat populations of E coli in a matter of seconds, leading to a new ‘steady state’ that is presumably reached by all cells within these populations [37, 38]. It is reasonable to assume that this has occurred in both the MDS42 and MG1655 chemostat populations in the present study. However, our screen enabled us to select individual mutants from these populations, and their growth rate phenotypes varied, suggesting that we were able to observe additional changes to this phenotype that deviates over and above the new steady state. Although our growth rate distributions (Fig. 2) suggested a high turnover of phenotypes, brought on by an elevated mutation rate, the fitness effect of mutations, and whether they could lead to adaptation were not clear. For the MDS42 mutants the first 14 days saw a wide distribution of slow growth rates, an indicator that mutations were likely to be deleterious, but insufficiently so to see them washed out of the chemostats quickly. Beyond 14 days a smaller range of growth rates was observed, which could be indicative of either a decrease in the accumulation of new mutants, or the possibility that the sustained high mutation rates led to the acquisition of increasingly deleterious phenotypes. This is plausible given that IS elements have been deleted from the MDS42 strains. Previous reports have suggested that the systematic deletion of IS elements hinders the ability of these multiple deletion strains to evolve in that acquired mutations, even with a detrimental effect on growth, are not fatal to the cell [15, 24]. Therefore, even in instances where high mutation rates are sustained, new mutations of varying effects continue to accumulate rapidly in MDS42 populations, which over time could make the population more susceptible to genetic drift or purifying selection events.
In contrast, the range and mean of the distribution of mutant growth rates in the MG1655 strain appear to fall after just 7 days in batch monocultures. In addition, by day 14, 94% of mutants would not grow in monocultures and those that did, grew very slowly, an indication of deleterious mutations. Yet surprisingly, mutant accumulation in the chemostats continued to rise over the 21 days. Furthermore, in mixed batch cultures, mutants appeared to thrive, with a growth rate approximately four times faster than when grown individually. So it appears that for the strain with no genome reductions, MG1655, the mutants that survive in the chemostat are no longer exploiting the same niche as the non-mutants; they do not grow in the original media. Their ability to thrive as a cohort suggests that in these strains, acquired mutations have delivered cross-feeding phenotypes. Cross-feeders metabolise by-products from other individuals in the population [39, 40]. Therefore, in a chemostat, a limited but renewed source of glucose could establish a hierarchy whereby a percentage of the population utilises this glucose, and produces byproducts that could, in turn, feed other members of the population, allowing cross feeders to adapt. Multiple byproducts could open up a multitude of niches, leading to the adaptation of several putative beneficial mutations. This could explain the weak, but present, correlation between mutant accumulation and deviation from steady state in the MG1655 chemostats, which suggests accumulating mutations were causing an increase in cell density. This correlation was absent in the MDS42 reactors.
Overall, our study showed that stress increased mutation rates in both reduced and non-reduced E. coli strains. The MDS42 strain has been engineered for stability and was intended for a closed and controlled biotechnological application [6]. Although the current study shows that the elevated mutation rate was the result of metabolic stress that is typical of an open environment, it stands to reason that closed systems would behave similarly when under any stress. Previously documented evidence from lab evolution studies using E. coli show that similar to any other stress, glucose-limitation in a chemostat elicits the general stress response affecting genes such as rpoS, a master regulator of this stress response [27]. This stress-response can also down-regulate the DNA mismatch-repair machinery via mutations in mutS and mutY which ultimately leads to an increase in background mutation rates [21, 22, 26]. Although not specifically assayed in the current study, these mutations are very likely to have occurred in our populations too. Moreover, chromosome replication times have been observed to change in stressed chemostat populations in association with altered expression of genes involved in DNA replication, repair, and recombination [41]. Here too, these genes are triggered by and control a multitude of processes involving DNA integrity and hence could lead to an increase in mutation rates if mis-expressed.
Given that stress can lead to mutations via so many different genetic pathways, it is reasonable to assume that a stress of any type, even in a closed system could lead to an elevated mutation rate. This study also found that the adaptive effect of mutations was difficult to predict. We observed a different effect in populations from each strain, neither of which would be ideal for a biotechnology. For example, the cross-feeding mutants that emerged in the MG1655 populations could produce unwanted byproducts. Moreover, if embedded in an environmental biotechnology, the adaptations could lead to the engineered microorganisms becoming bio contaminants. In the MDS42 strain, the deletions of IS elements appeared to initially ‘dampen’ the effect of mutations a potential advantage for a biotechnology, provided that the process was completed in a few days. For longer-term applications, or for proposed environmental biotechnologies, further chassis modifications would be required.