Abstract
A tantalizing question in microbial physiology is the inter-dependence and evolutionary potential of cellular stress response across multiple environmental dimensions. To address this question, we comprehensively characterized the cross-stress behavior of wild-type and evolved Escherichia coli populations in five abiotic stresses (n-butanol, osmotic, alkaline, acidic, and oxidative) by performing genome-scale genetic, transcriptional and growth profiling, thus identifying 18 cases of cross-stress protection and one case of cross-stress vulnerability. We identified 18 cases of cross-stress protection and one case of cross-stress vulnerability, along with core and stress-specific networks. We tested several hypotheses regarding the effect of the stress order, stress combinations, mutation reversal and varying environments to the evolution and final cellular fitness of the respective populations. Our results argue of a common systems-level core of the stress response with several crucial implicated pathways that include metal ion binding and glycolysis/gluconeogenesis that is further complemented by a stress-specific expression program.
Introduction
Regardless of their complexity, organisms have developed a rich molecular and behavioral repertoire to cope with environmental variations in order to maintain homeostasis and cellular function. The term stress is used to describe conditions where environmental parameters differ substantially from an organism’s optimal conditions for growth and for bacteria it has been an active area of research for decades 1, given its industrial and medical importance. Most studies have focused on responses to single stressors, such as pH 2-4, temperature 5-7, oxidation 8,9 and UV 10,11, thus providing an important insight on stress-related cellular responses and their underlying mechanisms. More recently, several studies have investigated cases of stress combinations and particularly of cross-stress protection, where exposure in a given stressor confers a fitness advantage against an exposure to a second stress 12-23.
Evidence of cross-stress behavior has been mostly circumstantial so far and documented throughout the microbial kingdom. Early work on glucose- and nitrogen-starved Escherichia coli cells showed increase survival rates after heat shock or hydrogen peroxide-mediated stress when compared to non-stressed cells 24, with the alternative sigma factor rpoH a crucial link for heat shock protein production during starvation stress 25. E. coli cells adapted to high ethanol concentrations had decreased growth under acidic stress 26 and high temperature environments induce a similar transcriptional program to that observed under low oxygen 27. Pre-adaptation to elevated temperature can reduce both cell death rate and mutation frequency caused by hydrogen peroxide in Lactobacillus plantarum 28. Osmotic stress was found to confer inducible heat tolerance in Salmonella typhimurium 15, while trehalose synthesis, which is important for osmoprotection, was shown to feature a heat-inducible component that increases heat tolerance of osmo-adapted Salmonella enterica cells 14. More recently, gene-deletion libraries were used to investigate hydrogen peroxide (H2O2) tolerance in yeast29.
The emergence of cross-stress behavior and the impact of the environmental correlation-structure during its evolution remain tantalizing questions that need to be further investigated. In a recent study, the fitness of E. coli populations in five stressors was measured after adaptation for 500 generations under various other stressors and cross-stress dependencies were found to be ubiquitous, highly interconnected and emerging within short timeframes 30,26,28. While studies like that provide valuable insights on the evolutionary potential of cross-stress behavior, our knowledge of static cross-stress dependencies as well as the effect of specific environmental characteristics to their dynamic, evolutionary potential remains limited. Environmental structure is known to be important for the trait evolution during periodic fluctuations 31-33, co-evolution in microbial communities 34 and can influence the rate and direction of evolution. Additionally, cross-stress responses may be dependent on the type of co-evolution or the order of exposure to the pair of stresses, while it is not clear if and in which cases, evolved mutants revert to wild-type phenotypes in the prolonged absence of the stressor.
In this study, we have constructed a comprehensive map of cross-stress behavior by measuring the population fitness in pair-wise stress combinations for five stressors (osmotic, oxidative, alkaline, acidic, n-butanol) that are biotechnologically and biologically important 3,9,35-37. By performing genome-wide transcriptional profiling for each case, we identified key genes and pathways that are implicated in this phenomenon (Figure 1A). We then compared the observed cross-stress response arising from a short-term exposure to sequential combination of stresses to the response resulting from a longer term evolution of several E. coli lineages over 500 generations under the respective stressors. We identified the genetic basis of these differences by genome-wide re-sequencing of the mutants. Several hypotheses related to the stress order, stress combinations, mutation reversal and evolution in varying environments were tested through further evolving cells already adapted to n-butanol and osmotic to a total of 1000 generations under the respective conditions (Figure 1B). Fitness assessment and re-sequencing of the evolved strains provide insight on the genetic basis of acquired stress resistance and cross-behavior, while functional analysis highlight pathways that play a central role in this phenomenon.
Results
Comprehensive map of the cross-stress behavior in E. coli
We tested the cross-stress behavior of E. coli MG1655 cells in all the possible pair combinations across five stressful environments (n-butanol, osmotic, acidic, alkaline and oxidative, four biological and two technical replicates). Out of the 25 possible combinations, 18 pairs (72%) have significant cross-stress behavior which is in accordance with the presence of a general stress response in E. coli (Figure 2A, Supplementary Table S2, Supplementary File S1). The highest cross-protection was observed when bacteria were exposed to acidic stress prior to n-butanol stress (DF = 1.15 ± 0.04, p-value < 1.46·10-3). We only detected one significant case of cross vulnerability, in the case of bacteria moved from oxidative to acid stress (DF = 0.97 ± 0.01, p-value < 1.91·10-4). We found a significant cross-protection effect between acidic media and high salt stress (DF = 1.10 ± 0.02, p-value < 5.17·10-4), which is a cross-stress behavior that previous studies found to be inconclusive in E. coli cells 16,38 but is known to occur in S. typhimurium 18. We also detected a cross-protection similar to that described for S. typhimurium and Listeria monocytogenes 13,15 between osmotic and oxidative stresses (DF = 1.14 ± 0.02, p-value < 6.41·10-3).
A cross-protection was observed in bacteria adapted to either high salt concentration or n-butanol and then exposed to oxidative stress (DF = 1.14 ± 0.02, p-value <6.41·10-3 and DF =1.14 ± 0.02, p-value < 0.02, respectively). A recent study has identified cross-stress vulnerability for these two combinations 39, however in that case, E. coli was evolved for 500 generations in the first stress, hence accumulating mutations that can alter its cross-stress profile with respect to the ancestral line. Indeed, we found that evolved and un-evolved cell lines had significantly different behaviors in both the original growth media and the stress-induced M9 media supplemented with 100 mM H2O2 (Supplementary Figures S1 and S2).
In both presence of n-butanol and oxidative stresses, a first exposure to 4 out of the 5 stresses gives a selective advantage to the bacteria. Pre-adaptation to other stresses except acidic provides a fitness advantage in the oxidative environment, partially analogous to what has been described for L. monocytogenes 19. Surprisingly, in acidic stress we have observed the least cross-stress protection from all the other conditioned strains and we also observed the sole case of cross-stress vulnerability in the case of the oxidative-adapted strains (Figure 2A). On the other hand, bacteria adapted in low pH are well-positioned to compete in all other stresses. To quantify the degree of positive or negative cross-stress behavior, we introduce cross-stress plots (Figure 2B) where the acid-conditioned cells possess the highest area of cross-stress protection (0.922), while oxidative-conditioned cells have the lowest area (0.411) that also harbors the sole case of cross-stress vulnerability.
Transcriptional profiles associated to cross-stress behavior
To analyze the underlying mechanism of cross-stress behavior we performed two types of transcriptional analysis. First, we performed a transcriptional profiling for each individual stressor at samples harvested 12 hours after exposure. Then we analyzed the transcriptional profiles after exposure to the second stressor (24 hours) to identify the differently expressed (DE) genes in the various cross-stress samples (Supplementary Tables S3-5, Supplementary File S2).
In the case of the genes differentially regulated in each stress, we identified 41, 203, 111 and 21 DE genes (p-value < 0.05 after Bonferroni correction) for n-butanol, osmotic, oxidative and acidic stress respectively. For each stress we merged results of the four different analysis (Supplementary Figure S3) and then data for each stress was used to make a new analysis that identifies DE found in more than one stress (Figure 3A). Interestingly, in acidic environments 18 of the 21 DEGs found (86%) overlap with other stresses that correlates well with the fact that this specific environment has the highest cross-stress area.
In osmotic and n-butanol stresses there is an under-expression of two operons, nar and hya, both reported to be over-expressed in anaerobiosis 40,41. The hya operon was also down-regulated under oxidative stress when compared to its expression under control conditions. The operon encodes for the synthesis of a hydrogenase that is implicated in pH stabilizing responses 41. The gadE, gadB genes from acidic resistance system AR2, the adiA gene from AR3 and the hdeAB acidic response 42-45 were found to be under-expressed under the osmotic and n-butanol stressor when compared to their expression in the absence of a stressor. As expected, the acidic resistance gene gadB is overexpressed under acidic stress conditions. We have identified both a common core of DEG for stress combinations and top ranked genes for individual stresses (Supplementary Table S6, Supplementary File S2). In osmotic stress, proX that belongs to the operon that encodes for osmoprotectant transportation is significantly overexpressed 46,47, while in oxidative stress we have high expression of the yhjA and pfkA genes. The first is a the oxyR-regulated peroxidase 48 while the second is of unknown function and we have recently showed that it is implicated with H2O2 sensitivity 49. Surprisingly, the highest ranked DEGs for n-butanol are the whole hya operon and the gadB gene, which are both related to acidic resistance. Indeed, culture in n-butanol was found to lower the pH to 5.9 after 12h, which explains the up-regulation of the acidic-stress cluster and the resulting cross-stress protection that was observed in this study.
To measure transcriptional changes in the profiles for both cross-stress protection and vulnerability cases, we performed RNA-Seq on four stress combinations: cells adapted to osmotic stress and then exposed to n-butanol (cross-stress protection) and cells adapted in oxidative stress that were subsequently exposed to acidic stress (cross-stress vulnerability) and their complementary pairs (from n-butanol to osmotic stress and from acidic to oxidative) to identify whether the observed difference in Darwinian Fitness can be attributed to the enactment of a dissimilar transcriptional program. Two controls were used as reference: cells grown for 12 hours in M9 and then 12 hours in the second stress and a population that has been exposed to the second stress continuously for 24 hours (Figure 3B). In all the analyses DEGs are similar when compared to both references, except from acidic to oxidative stress (Supplementary File S2). When bacteria were exposed to n-butanol stress after being exposed to osmotic stress, there is only a small variation of expression compared to the bacteria that have been only exposed to n-butanol (9 DEGs only) showing that exposure to high salt conditions triggers a response similar to the response of bacteria exposed to n-butanol and no rewiring is needed giving the bacteria an advantage. Most of these 9 DEGs are acidic-response genes as found in the n-butanol samples like gadB and gadE (Supplementary Table S7, Supplementary File S2). In contrast, when bacteria that has adapted in n-butanol face osmotic stress, more than 100 genes are differentially expressed when compared to the expression under osmotic stress exposure only. A similarly large differentially expressed gene pool is observed in the case of cells exposed to oxidative stress prior to acidic stress, which also is the only cross-vulnerability observed, where the list of DEGs goes over 750 genes, arguing for the cell’s need to enact a different expression program than the one used in the first stress in order to survive, which concords with a fitness disadvantage. In the 20 most significant DEGs (Supplementary Table S8, Supplementary File S2) there are several up-regulated genes previously described to be differentially expressed different stresses like acidic (yjeI, 50), manganese (hflX, 51), cobalt (iscR, 52) cadmium (tyrA, 53) and oxidative (pfkA, 49). In the inverse case, there is a cross-stress protection from acidic to oxidative and transcriptional profiling shows only 10 DEGs compared to the acidic grown bacteria that contains acidic related genes (hdeB, gadB, hyaB).
Dissecting the memory imprint after adaptation to sequential stress combinations
Osmotic and n-butanol adapted cells demonstrate strong cross-stress protection both in the case of a short 12h exposure (Figure 2A) and adaptation after 500 generations 39. However, contrary to past results on adapted cells, where cross-protection has been found to be symmetrical, only bacteria exposed to n-butanol are protected from osmotic stress which argues for the existence of directionality of this phenomenon. To assess this hypothesis, we evolved osmotic and n-butanol adapted strains for an additional 500 generations in five independent experiments, all combinations of these environments, to further address questions related to cross-stress behavior after sequential stress adaptation (Supplementary Table S9). Growth rates of all the populations in media-only, osmotic stress and n-butanol showed that there was no growing defects after evolution (Supplementary Figures S4-6, Supplementary Table S10, Supplementary File S3).
The first hypothesis we tested is whether exposure to two stresses sequentially provides an advantage in an environment where both are present. We performed competition assays in an environment with both stresses (0.3mM NaCl and 0.6% n-butanol) between populations evolved for 1000 generations in a single stress (O1000, B1000) against those evolved in both stresses sequentially for 500 generations (O500B500, B500O500). Results show a significant fitness increase of the strains that have been exposed to both stresses (O500B500 vs B1000, DF = 1.16 ± 0.08; B500O500 vs O1000, DF = 1.24 ± 0.12) (Figure 4A, Supplementary Table S11, Supplementary File S4). This memory effect is present long after the selection pressure for this particular stressor was removed (500 generations) and it can be attributed to the genetic background that has been acquired during the adaptation in the first stressor. In addition, this fitness increase is sufficiently large to overcome the additional fitness increase that is observed in the O1000/B1000 strains when compared to their O500/B500 counterparts due to their prolonged stress adaptation (DF = 1.12 ± 0.03, p-value < 7.15·10-3).
Recent theoretical results show that the rate of evolution can be increased by guiding populations from states that are correlated and have increasing complexity 54. To challenge this notion in the case of osmotic and n-butanol adaptation, we competed populations evolved sequentially in both environments to those evolved in only one stress. In both cases the population evolved for a prolonged duration in a single stressor had similar or higher fitness than populations evolved in the two stressors sequentially (O500B500 vs. B1000, DF = 1.03 ± 0.02, p-value < 0.58; B500O500 vs. O1000, DF = 0.87 ± 0.01, p-value < 7.54·10-11). Concomitantly, we investigated if the order in which stressors are introduced play a role on the final fitness. We found that order does not have a significant impact to the fitness of the final populations when it is measured under competition in either stress (O500B500 vs. B500O500, DF = 0.99 ± 0.11, p-value < 0.99 in n-butanol stress; DF = 0.99 ± 0.010, p-value < 0.88 in osmotic stress).
We further probed on the environment’s impression on the genetic blueprint by investigating whether mutations are reversed when the stressor is removed. As such, populations that were adapted to environments for 500 generations with either n-butanol or osmotic stress (B500 and O500 populations, respectively) were competed to those where the initial 500 generations of stress-adaptation is followed by another 500 generations where the stressor is removed (medium-only environment; B500G500 and O500G500 cell lines). In the case of n-butanol, there is no significant difference in the Darwinian fitness when populations are competed under n-butanol stress (B500 vs. B500G500, DF = 0.99 ± 0.01, p-value < 0.34), while in the case of osmotic stress, the difference is profound in competition under osmotic stress, as beneficial mutations that are advantageous in osmotic stress have been lost (O500 vs. O500G500, DF = 0.79 ± 0.01, p-value = 8.30·10-6). In both cases, growth in media without the presence of stress was not affected (DF = 1.05 ± 0.04, p-value = 0.39; DF = 0.97 ± 0.03, p-value = 0.44).
Finally, to investigate the effect of alternating stresses to the evolutionary trajectory of each population, we evolved the O500 and B500 populations to alternating environments of osmotic and n-butanol stress (50 generations per environment, 10 switching events total; O500BO50 and B500OB50 cell lines). These populations have been competed with O500B500 and B500O500, respectively, in three distinct environments: n-butanol, osmotic stress and a varying environment with 8 hours for each stresses (Figure 4A). We did not observe any difference in the growth curves and competition assays between any of the populations grown in varying and static environments (B500OB50 and O500BO50 vs. B500 and O500 respectively; Supplementary Table S11).
Acquired genetic mutations and their effect to phenotypic fitness during short-term evolution
Previous studies on the effect of single stress adaptation to cross-stress behavior have been focused on the identification of mutations on one cell line over short timeframes 39. To further elucidate the genetic basis of the acquired stress resistance and cross-stress behavior, we first sequenced four cell lines for each of the four of the stressors (osmotic, n-butanol, acidic and oxidative; alkaline stress was excluded because of media incompatibility) and the control environment (4 biological replicates, 5 environments, 20 lines total), where we have identified a number of loci that are mutation hotspots over two or more cell lines (Figure 4B, Supplementary Table S12).
Across all lines, two mutations were found to be synonymous, ynfL (G500) and cydD (O500) (Supplementary Table S13). The predominant mutations are in the intergenic zones rph:pyre (G500, P500, H500) and fes:fepA (O500, B500), as well as the sigma factors rpoB and rpoC (G500, O500, B500, H500). Interestingly, both fes and fepA are involved in siderophore enterobactin production that has recently been found to be linked with oxidative stress and M9 growth 55. Although its function is unknown, mutations in rph:pyrE have also appeared in a previous study in evolution on lactate 56. Both rpoB and rpoC encode for RNA polymerases and their mutations are very usual in evolution, more specifically mutations in rpoB have been linked to an increased evolvability and fitness 57 and mutations in rpoC to bacterial growth optimization in minimal media 58 and metabolic efficiency 59. We also found mutations in other genes already described like pykF in G500 which mutation is known to be beneficial for growth in M9 57, marC and acrAB for resistance in n-butanol 39,60,61, proV in osmotic stress 39,46,47; evgS and rpoD in acidic stress 39,42,62 and katG, oxyR, rsxD and ccmD for resistance in oxidative stress 39,63,64. Interestingly, in oxidative stress we find mutations in the yagA and yncG genes, which is consistent with a previous finding that a deletion between argF-lacZ (that includes yagA) confers a high hydrogen peroxide resistance 65. In addition, yncG is homologous to Glutathione S-transferases known to help in the defense against oxidative stress 66. Other mutations with unclear importance in the stress where they appeared that were found are ybcS, lon and prc in G500; promoters of yccF, ychF and pgi and genes rob and relA in B500; yijO in 0500; lon in P500; and motB, yigA and the intergenic zone panC:panB H500. The mutations in the ybcS, prc, yccF and yigA genes are of unknown function, while others have known, but unrelated functions. For example, lon is a DNA-binding protease that degrades abnormal proteins 67, ychF is a ribosome binding catalase 68, pgi is an oxidative stress induced gene 69, rob is a transcriptional activator involved in antibiotic resistance 70,71, relA synthesizes ppGpp, an alarmone active under aminoacid starvation conditions 72 and motB is part of the flagellar motor 73.
We performed a similar re-sequencing analysis in the case of populations that have been evolved for a total of 1000 generations in a single or multiple sequential stress combinations (sequencing of multiple clones per population; Supplementary Figure S7, Tables S14-16). Four mutations were found to be synonymous and hence neutral, paaH (B500O500), flgK (O500B500), fixB in (O500G500) and ynfL (O500BO50). As shown in Figure 4B, the most common mutations are in the fepA:fes and rph:pyrE intergenic regions (8 and 7 cell lines, respectively) and mutations in the rpoB and rpoC genes (7 and 6 cell lines, respectively; Table 1). Many of these targets are mutation hotspots (Supplementary Table S16), for example in the case of rph:pyrE and fepA:fes, parallel mutations have been detected in 9 positions that span 140 nucleotides (Figure S8) and 3 positions over 22 nucleotides (Figure S9) respectively. Populations of the B500 parental population contained several genes present in more than one population, marC, acrA, acrB, relA, glyXVY, with valZ:lysY. marC and acrAB have been previously described to be involved in n-butanol resistance 39,60,61 along with nagA. In contrast, populations evolved from the O500 background have only two common mutagenized loci present: proV described to be involved in osmotic resistance 39,46,47 and nagA present in osmo-tolerant cells 35.
To decipher the genetic basis of this reversal, we compared the sequences of the B500 and O500 clones (Figure 4A) to those from B500G500 and O500G500 populations, respectively (Supplementary Figure S7). Cells that were first evolved in n-butanol and then adapted to stress-free media for 500 generations lost the rpoB mutation while acquired mutations in pykF, gltD and gltB genes. Similarly, cells that were first evolved in osmotic stress and then to stress-free media lost the mutation to the osmoprotectant gene proV and acquired mutations in the pyrE:rph tap, yeiP and pykF clusters. Although its function is unknown, mutations in pykF gives an advantage for growth in M9 57. Both gltD and gltB encode genes for glutamate synthesis 74, tap produces a methyl-accepting protein 75 and yeiP is an elongation factor.
To associate genotype-to-phenotype, we performed growth experiments of single-knockout mutants 76 in medium-only, n-butanol and osmotic stress (Supplementary Table S17, Supplementary Figure S10, Supplementary File S3). Mutants in pyrE and fes cannot grow in any of the conditions tested, showing that the mutations found in their promoter affect the growth in M9 and are not involved in the stress adaptation. While mutation in fepA affects growth in all media, it provides a higher maximum growth rate in osmotic stress (Supplementary Table S17, media-only μmax = 0.011 ± 0.000; n-butanol, μmax = 0.014 ± 0.001; osmotic, μmax = 0.020 ± 0.000). As previously described, ΔpykF enhances fitness in media-only 57, ΔacrA and ΔacrB in n-butanol 39,60 and ΔproV in osmotic stress 46,47. Other mutations that have a positive effect are ΔrelA and ΔsufE in media-only, ΔyjbI in n-butanol and ΔacrA in osmotic. Previous results of repaired mutations under the mutant background elucidated the effect of acrA mutations in osmotic stress 39. Cross-stress trade-offs where found in the case of ΔnagA and ΔsufE that have a negative effect under osmotic stress. In constrast, no mutation was found to have a negative effect in growth under the absence of any stressor, or in presence of n-butanol.
Functional and network analysis of cross-stress behavior
Re-sequencing and transcriptional profiling data were compiled in order to produce a network representation of stress resistance in E. coli (Figure 5A, Supplementary Figure S11). Network and functional analysis was also performed for each individual stress (Supplementary Figures S12-S15). Overall, enriched gene ontology clusters are related to motility, sulfur metabolic process, translation, DNA replication and cellular respiration, among others (Figure 5A and B). Interestingly, the glycolysis/gluconeogenesis, response to drug genes and cellular respiration are involved in n-butanol response, with an pykF mutant from this pathway being unexpectedly fit, as it has the highest growth rate in these conditions. The importance of the fes and fesA genes of the Metal Iron Binding pathway is profound in the case of osmotic and oxidative stress and similar results have been obtained in the case of fepA iron transporter, in n-butanol, osmotic and acidic stress, highlighting the importance of metal transport in stress response. In osmotic stress, several translation and transcription pathways are found to be involved, while in acidic stress regulation of transcription and cell cycle clusters are enriched. The metal binding motif also includes a differential expressed gene, yhjA, that encodes for a cytochrome C peroxidase regulated by FNR and OxyR 48. This gene is been implicated in all stress responses and hence is an excellent target for further characterization. In this stress we also find motility genes involved in bacterial chemotaxis. In our analysis, a core network was identified with DEGs and mutations from every stress (33% n-butanol, 19% osmotic, 15% acidic and 33% oxidative). The fact that the members of the central clusters are all implicated in the acidic and osmotic stress correlates well with the previously shown cross-stress plots in which we showed that exposure to acidic and osmotic stresses have the higher cross-stress protection.
Discussion
A general observation that stems from our work is that exposure of a bacterial population to one stressor is generally beneficial, as 18 out of 25 cases tested in this work showed cross-stress protection with an average Darwinian Fitness of 1.08 ± 0.01 (p-value < 5.52·10−8). Interestingly, we also found the first case of cross-vulnerability when bacteria adapted to oxidative stress face low pH suggesting that some stress combinations can be used for a better sterilization. By performing transcriptional profiling we found that the number of differentially expressed genes across stresses is a good predictor of the underlying cross-stress behavior. Indeed, the case of cross vulnerability has the higher number of DEGs (>750), followed by the neutral cross-stress (>100) and finally both cases of cross-protection have a small amount of DEGs (9 and 10). Although the results here support this hypothesis, cross-stress protection can be also dependent on the DEGs of the first stressor as we find several genes with unknown or unrelated function, including the putative transcriptional repressor rpiR, the yfi operon and genes pqqL and ybaT that are present in all stresses but acidic (Fig. 3, Supplementary file S2). These genes constitute excellent targets for further experimentation to understand the mechanism under which they affect single stress and cross-stress behavior. Additionally, we identified several genes known to be differentially expressed under anaerobic conditions, which provide a clear link between genes induced by anaerobic respiration and stress resistance 77.
There is substantial similarly of the cross-stress behavior between n-butanol and osmotic stress, as well as a clear dissimilarity of these responses to that of oxidative stress with and without evolution. Analysis of the mutations found in the 4 clones sequenced (B500 and 0500, Figure 4A) discovered two genes in the n-butanol evolved populations that are also involved in oxidative stress that might be related with this differential behavior, the ychF catalase 68 and the pgi gene that are induced under oxidative conditions 69. Although not directly related to oxidative stress, in these clones we also found a mutation in the rob gene involved in antibiotic resistance 70. Unpredictably, in the O500 clones only one mutation is not related to the media, a mutation of the gene yijO which is of unknown function and remains to be investigated. Another explanation of the cross stress vulnerability is the fepA:fes mutation that, although it is expected to be related to the use of M9 as evolution media, it is only present in the B500 and O500 populations. The corresponding proteins belong to the enterobactin operon and have been recently implicated in growth under oxidative stress in M9 media 55 and thus its mutation might have a significant effect in M9 with oxidative stress.
If we are to predict where mutations will occur under evolution in a specific stress, it is reasonable to assume that the genes that are differentially expressed in that given stress will be likely candidates for polymorphisms or silencing. Comparison of the transcriptional profiles after short-term exposure and re-sequencing of evolved populations after 500 and 1000 generations argue that, in general, this hypothesis is not correct, as differential expression under exposure in a single stressor is not a proxy for mutation events during evolution in that stressor. Only in the case of osmotic stress we observe a mutation in the gene proV after evolution, which is in the same operon as the proX gene that was found to be over-expressed after the 12h exposure. Interestingly, after 1000 generations under n-butanol stress a mutation in relA appears and by network analysis we show that this gene is very close to several DEGs described in this stress like gadE, mdtE and mdtF. A similar case exists in oxidative stress where we detected a mutation in the intergenic region between panB and panC, with the network analysis showing that these genes have several interactions with other oxidative stress specific DEGs, such as dnaJ, hypD and lolB.
The evolved cell lines provide an interesting view of the evolutionary trajectories under the various stress combinations and it shows that their diversity is environment-specific. As shown in Figure 4B, cell populations that evolved under osmotic stress share few genetic mutations while populations evolved under n-butanol stress share many. Interestingly, despite the already described n-butanol related mutations, acrA, acrB and marC, we also found point mutations in several regions between tRNAs, metW:metV, valZ:lysY, glyC:glyY and valT:lysW. Mutations in four different spots argue against this being a random effect, which in turn implicates the tRNA production as a factor of n-butanol stress resistance. An expected result is the high abundance of mutations in the fes:fepA genes in all the populations as we already described this mutations in the parental population, B500 and O500. These mutations have been already described in similar scenarios and recently the involvement of the fep operon in growth in M9 39,55,56. Surprisingly, another widely present mutation, pyrE:rph, is only present in the population evolved for 500 generations in a stress-free M9 medium, which argues for the fes:fepA mutation to be stress-related. We can track the emergence of the most common mutations by the number of polymorphisms detected in each mutation, for example the fes:fepA mutations appears in three different positions (Supplementary Figure S9) for the B500 and O500 populations. In the case of rph:pyrE, we have 9 distinct mutation positions (12 different mutations Supplementary Figure S8) arguing for the emergence of these mutations in the last 500 generations of the evolution. In addition, based on our results, we cannot conclusively dismiss or confirm the guided-evolution hypothesis although the O500B500 and B500O500 cell lines were found to have similar fitness to the 1000 generation populations, B1000 and O1000, in n-butanol and osmotic stresses, respectively (Figure 4A). This was surprising as the O500 and B500 have a significantly lower fitness with respect to their 1000 generation counterparts.
Finally, our last experiment involved fitness in a varying environment. More experimentation and in larger scale is needed to state with high confidence the magnitude and occurrence of this phenomenon. A recent work was also inconclusive regarding this phenomenon in S. cerevisae 31 with five times shorter cycles (approximately 10 generations per cycle). We also tested the hypothesis that shorter lag phases, instead of better fitness, should be observed under evolution in varying environments (New et al, 2014). Comparison of the μmax of the varying environments population (Supplementary Table S10) shows that the O500BO50 general fitness is better than the O500B500 in osmotic stress (0.064 vs. 0.0525) but not in n-butanol (0.041 vs. 0.0375). In the case of the B500OB50 populations, a higher μmax in osmotic stress than the B500O500 (0.0575 vs. 0.052) and a similar growth in n-butanol stress (0.039 vs. 0.043) was observed.
Although cross-stress behavior has been extensively studied in the past years, with many such cases related to food safety 78 and human health 79 this is one of the first studies where genome-wide profiling and a systems biology approach was used to create a comprehensive mapping of static and evolved cross-stress responses. This approach allows us to investigate the systemic response and molecular underpinnings of each condition combination, with the potential to inform new sterilization techniques and more efficient methods in clinical decision support 80.
Acknowledgements
We would like to thank Athanasios Tsoukalas for his help with identifying the aminoacid effect of mutation events. This work was supported by NSF awards 1244626, 1254205 and DoD Army Research Office award W911NF1210231-0.
Author Contributions
VZ performed cross-stress experiments, growth curves, strain characterization, and library preparation for genome sequencing and RNA-Seq. SQS performed the strain evolution and performed competition assays for validating the fitness of the populations. MK analyzed the sequences and the DEGs. NR performed the libraries for the X500 clones. VZ and IT evaluated data and prepared the manuscript. IT conceived of the study and supervised all aspects of the project.
Conflict of Interest
The authors declare no competing interests.
Materials and methods
Bacterial strains and culture conditions
E. coli MG1655 and MG1655 ΔlacZ strains were used in all experiments of this study. The inclusion of ΔlacZ mutants allowed us to perform competition assays using X-Gal and IPTG staining. The neutrality of the ΔlacZ mutation was confirmed in all environments in a previous work of the laboratory 39. Minimal M9 salt medium with 0.4% (w/v) glucose as carbon source was used for cross-stress behavior competition assays. The stresses used are osmotic stress (0.3M NaCl), acidic stress (pH 5.5), alkaline stress (pH 9.0), oxidative stress (100 mM H2O2), n-butanol stress (0.6% n-butanol), and control (no-stress).
Cross-stress behavior analysis
Two competing populations (E. coli MG1655 and E. coli MG1655 ΔlacZ) were pre-adapted to M9 overnight and inoculated at an approximate 1:50 ratio at a starting OD600 0.004 in M9 with one stressor and only M9 for reference. After 12 h of growth, the same quantity of the strain adapted to stress and the strain grown in M9 were diluted in the second condition media (approx. 1:100 dilution) and competed for additionally 12 h. Samples were taken at 12 (time 0) and 24 hours (time 1). Four independent biological replicates each with two technical replicates were performed for each individual competition. Cell counts were determined on LB agar plates containing 0.25mM IPTG (Isopropyl β-D-1-thiogalactopyranoside) and 40 mg/ml X-gal (bromo-chloro-indolylgalactopyranoside). Plates were incubated at 37°C overnight. Darwinian Fitness (DF) was calculated as described in the Supplementary Methods section.
Gene expression analyses
Cells were harvested at time 0 and 1. In all, 3ml aliquots were harvested and mixed with 1.5 ml 5% Phenol/ethanol (v/v) and stored at 80C until use. RNA was extracted using an RNeasy kit (Qiagen) and after first- and second strand cDNA synthesis, cDNA was broken using Diagenode Bioruptor NGS. End repairing, A tailing, linker ligation and PCR enrichment was made using the KAPA Library Preparation Kit (Kapa Biosystems). Size selection was performed with Agencourt AMPure XP (Beckman Coulter). After quality control, libraries were sequenced by Illumina HiSeq 2500.The low-quality raw reads were trimmed using Trimmomatic (v0.30) with default settings. Trimmed reads were aligned on most recent reference genome of E. coli MG1655 by using TopHat (v2.0.10) coupled with bowtie (v1.0.0) 27,81. The identification of the differentially expressed gene was done as explained in the Supplementary Methods.
Adaptive Laboratory Evolution
Evolutionary adaptation experiments were performed in 10 ml volume in 125 ml shake flasks at 37C on an orbital shaker at 150 r.p.m. of an alternate of M9 with 85 mM NaCl (evolution media)media-only or with two stressors: osmotic stress (0.3M NaCl) and n-butanol stress (0.6% n-butanol). The OD600 of each culture was measured each day before the daily transfers to ensure that the estimated nine generations per day were reached (see Supplementary Methods).
Whole-genome re-sequencing and mutation discovery of evolved strains
Selected clones were grown on LB medium overnight and genomic DNA (gDNA) was isolated using a Wizard Genomic DNA Purification Kit (Promega) and sequenced as described in the Supplementary Methods.
Competition assays
Competing populations (E. coli MG1655 and E. coli MG1655 ΔlacZ) were grown under adaptive conditions overnight and inoculated at an approximate 1:1 ratio at a starting OD600 0.004. Samples were taken at 0h and 24h. Two independent technical replicates were performed for each individual competition. Cell counts were determined as described before, using LB agar plates containing 0.25mM IPTG and 40 mg/ml X-. Plates were incubated at 37°C overnight. Darwinian fitness (W) was calculated as described in the Supplementary Methods section. In the case of the population adapted to a fluctuating media, the populations competed in a fluctuating media that varied from osmotic to n-butanol every 8 hours.
Network analysis
We used TF-DNA binding data from RegulonDB 82and protein-protein interaction datasets 83,84 from the Bacterial Protein Interaction database 25 to construct a functional and regulatory network. We then mapped the results from our re-sequencing and transcriptional profiling analyses to build condition-specific sub-networks for each stress. Genes that connect either mutated or DE genes in paths with a path length of three or lower were also included in the network analysis. Modular organizations and community detection was performed on the resulting networks, through spin-glass model and simulated annealing techniques 85 using igraph R package 81.