Experimental evolution and phenotypic characterization of end populations
E. coli EcNR1, a derivative of E. coli K12 MG1655 harboring a λ Red prophage integrated at the bio locus, was evolved by serial passaging of six independent populations for approximately 500 generations on isobutanol spiked M9 minimal medium supplemented with 50 g/L carbon source and 0.25 mg/L biotin. An initial isobutanol concentration of 0.75% (w/v) (corresponding to approximately 75% growth inhibition) was used for all populations, providing strong selective pressure. Isobutanol concentration was gradually increased during evolution to maintain approximately constant selective pressure. Populations were evolved on two different carbon sources, with three populations evolved with 50 g/L glucose as the sole carbon source (designated glucose #1, glucose #2, and glucose #3, abbreviated G1, G2, G3) and another three populations evolved with 50 g/L xylose as the sole carbon source (designated xylose #1, xylose #2, and xylose #3, abbreviated X1, X2, X3). Glucose and xylose are important constituents of lignocellulosic feedstocks and are metabolized by different pathways; thus we explore adaptations in different metabolic contexts relevant to biofuel production . Cultures from each evolving population were periodically archived by cryopreservation, and phenotyped by measuring maximum specific growth rate (μmax, h-1) at various isobutanol concentrations using a microplate spectrophotometer.
All of the evolved populations show significantly improved fitness at high isobutanol concentrations relative to the parent E. coli EcNR1 strain (WT) (Figure 1A and 1B). Interestingly, the populations show divergent growth phenotypes. Clonal isolates from two highly tolerant populations, G3 (glucose #3 population) and X3 (xylose #3 population), were further phenotyped, revealing significant heterogeneity within these populations (Figure 1C and 1D). Three clones from G3 were capable of growth at 2% isobutanol in glucose media and two clones from X3 grew at 1.75% isobutanol in xylose media, representing 60% and 40% improvements in tolerance respectively, compared to WT (Figure 1C and 1D). Two representative clones with high fitness, G3.2 and X3.5, were chosen for further characterization.
Evolution often produces adaptations that show tradeoffs in relative fitness across different environments . To investigate specificity of adaptation, the fitness (relative to WT) of clones G3.2 and X3.5 at 0% and 1% (w/v) isobutanol was assessed on minimal glucose, minimal xylose, and rich LB media (Figure 2A and 2B). At 0% (w/v) isobutanol both G3.2 and X3.5 show improved fitness on xylose minimal medium and decreased fitness on LB medium, relative to WT (Figure 2A). At 1% (w/v) isobutanol, G3.2 and X3.5 show markedly improved relative fitness on both glucose and xylose minimal media, and decreased fitness in LB medium (Figure 2B). These results suggest that the two isolates characterized have accumulated adaptations to isobutanol stress specific to minimal media, and these adaptations appear to exhibit antagonistic pleiotropy in rich medium. This minimal-rich medium antagonistic pleiotropy we observed underscores the importance of carefully selecting evolution conditions. On the other hand, although G3.2 and X3.5 were evolved on glucose and xylose media respectively, neither of these strains appears to have developed carbon-source specific adaptations in 0% and 1% isobutanol environments. We further assayed fitness in glucose and xylose minimal media at higher isobutanol concentrations (Figure 2C). At 1.5% (w/v) isobutanol, we observed relative fitness trends suggesting greater specificity of adaptation for G3.2 and X3.5 to their respective carbon sources, but we could not substantiate that these differences were statistically significant due to the error bars in our measurements (Figure 2C). Interestingly, at 0% isobutanol X3.5 appears to have higher relative fitness than G3.2 in all media types tested (Figure 2A). ATP yield from xylose metabolism is lower compared to glucose metabolism, and we speculate that low ATP yield increases selective pressure for more energy efficient use of carbon sources . This may explain how adaptations to a low ATP yield substrate such as xylose could also be beneficial to growth on other carbon sources.
In addition to investigating specificity of adaptation to different carbon sources, we also examined the tolerance of G3.2 and X3.5 to various alcohols with potential for microbial biofuel production, including ethanol, isopropanol, and n-hexanol (Figure 2D). While all alcohols share the same general mechanisms of toxicity via chaotropic effects and interactions with membrane lipid bilayers, specific biophysical effects are known to vary with alcohol chain length . Molecular dynamics simulations and experiments with model lipid bilayers have demonstrated that long chain alcohols (≥ C8) tend to condense and stiffen lipid bilayers, while short chain alcohols (≤ C2) have opposite effects ; lipid bilayer interactions with intermediate length and branched alcohols (such as isobutanol) have not been well characterized. We examined the percent relative inhibition of WT compared to G3.2 and X3.5 (defined as , where μWT is the maximum specific growth rate of E. coli EcNR1, and μMUT is the maximum specific growth rate of G3.2 or X3.5) at 3.5% (v/v) ethanol, 2.5% (v/v) isopropanol, 0.5% (w/v) isobutanol, and 0.25% (v/v) n-hexanol; concentrations were chosen to correspond to approximately 1/2 of the minimum growth inhibiting concentration (MIC) on glucose minimal medium at 30°C. For all alcohols assayed, G3.2 and X3.5 displayed higher tolerance than WT; interestingly, the relative inhibition of WT increased with increasing chain length (hexanol > isobutanol ≥ isopropanol ≥ ethanol), indicating the adaptations to isobutanol stress may be selective to the effects of longer chain alcohols.
Genome resequencing of isobutanol tolerant clones
To identify the genetic bases of adaptation to isobutanol stress, we resequenced the genomes of highly tolerant clones from our evolved populations with the Illumina Solexa platform, using 36 base pair single-end or paired-end read configurations. 612 to 756 million base pairs (MB) of raw sequence was generated for each sequenced genome with single-end read configuration, with approximately four times as much sequence generated with paired-end reads. Coverage averaged approximately 125× and 500× for the 4.65 MB E. coli EcNR1 genome, using single-end and paired-end reads, respectively. Reads were mapped to the E. coli EcNR1 reference genome sequence using Novoalign v2.04.02 and MAQ v0.7; single nucleotide polymorphisms (SNPs) and short insertion/deletions (indels) were called from the consensus sequence [25, 26]. Larger indels were detected by examining coverage distribution. Unmapped reads were collected and de novo assembled using Velvet v0.7.51 to detect breakpoints near sites of structural variation (SV) .
We resequenced the genomes of G3.2, G3.6, and X3.5, three highly isobutanol tolerant clones from the evolution end populations (discovered mutations summarized in Figure 3 and Table 1; full mutation lists available in Additional file 1 and the reference genome sequence in Additional file 2). It was discovered that the G3 lineage had acquired a 19 bp deletion in mutL, a component of the methyl-directed mismatch repair system (MMR). MMR loss-of-function mutations lead to an approximately 100-fold increase in mutation rate, giving rise to the so-called mutator phenotypes . Subsequently G3.2 and G3.6 were highly mutated, having 48 and 64 mutations respectively, with 20 mutations in common between these two clones (Figure 3B and Table 1). To narrow down candidate mechanisms of genetic adaptation, we resequenced the genome of a non-mutator clonal isolate from generation 266 of the G3 lineage (G3.266.7) and identified 8 mutations in this clone (Figure 3B and Table 1). For X3.5, 11 mutations were revealed (Figure 3A and Table 1).
A total of 131 mutations were discovered across clones X3.5, G3.2, G3.6, and G3.266.7 (full list available in Additional file 1). 96 mutations were SNPs, 25 mutations were short indels, and 10 mutations were SVs. Most mutations occurred in the coding region of genes. The detected SVs consisted of transposon insertions (marC::IS1 in all sequenced isolates, glnE::IS186 in the G3 clones, and mdtJ::IS5::tqsA in X3.5), an approximately 10 kb deletion between gltB and yhcE in G3.266.7, and a 1688 bp deletion in the ycfK gene of the e14 prophage in G3.266.7. Mutations were found in diverse genetic loci representing many cellular processes. BiNGO (Biological Network Gene Ontology tool) was used to assess any overrepresented Gene Ontology (GO) terms in the full mutation set, but the only statistically significant finding was an enrichment of membrane proteins (corrected p-value = 7.23 × 10-3), with a borderline significant enrichment of RNA helicases (corrected p-value = 7.22 × 10-2) .
Comparison of the genotypes of X3.5, G3.2, G3.6, and G3.266.7 reveals a number of parallel genotypic adaptations. In particular, mutations in rph, acrAB, marC, mdh, and the gatYZABCD operon were found in all of these clones (Table 1). E. coli K12 MG1655 (the parent strain of E. coli EcNR1) has a 1 bp deletion in the rph-pyrE operon, resulting in reduced levels of orotate phosphoribosyltransferase (the product of pyrE) and subsequently suboptimal pyrimidine biosynthesis levels . Thus restorative mutations are commonly observed in rph-pyrE during experimental evolution studies with E. coli K12 MG1655, and are general adaptations to growth on minimal media. The AcrAB proteins are components of the AcrAB-TolC multidrug efflux pump, a membrane transporter which translocates a wide range of substrates out of the cytoplasmic membrane and periplasmic space; efflux via the AcrAB-TolC complex has been previously identified as an important mechanism of tolerance to organic solvents such as toluene, immediately suggesting a possible role for acrAB-tolC in isobutanol tolerance . Possible links to isobutanol tolerance are not as obvious for marC (a predicted membrane protein of unknown function), mdh (NADH dependent malate dehydrogenase), and the gatYZABCD operon, which encodes proteins involved in galactitol transport and catabolism.
To investigate possible parallel genotypic adaptations in our other evolved lineages, the acrAB operon, tolC, and mdh were sequenced in 8 clonal isolates from each of the evolved endpoint populations (Table 2). The marC locus was also sequenced in each endpoint population; examination of PCR products revealed indel mutations (discernable by product size) at near 100% allele frequency, allowing for whole population samples to be sequenced (Table 2). We also sequenced the post-transcriptional regulator hfq in our endpoint populations since an hfq mutation was found in G3, and modulation of hfq has been observed as a common mechanism of adaptation in other experimental evolution studies. rph and gatYZABCD were not investigated further since rph-pyrE adaptations have been characterized in previous works, while the relatively large size of the gatYZABCD operon was prohibitive for Sanger sequencing.
acrAB mutations were discovered in X1, X2, X3, G1, and G3 populations (Table 2). Each population fixed only a single mutation in acrA or acrB, and allele frequency was near 100% (8/8 clones) except for G1, which had an allele frequency of approximately 25% (2/8 clones) and X3, with an approximate 50% allele frequency (4/8 clones). We did not detect acrAB mutations in the G2 population, which intriguingly had the lowest fitness out of the six endpoint populations. tolC mutations were not detected in any of the populations. The fixation of acrAB mutations in five out of six independent populations suggests strong selective pressure and parallel evolution at this locus. Mutations affected amino residues at a variety of positions in the protein structure (Figure 4A). The acrAB mutations acquired in the isobutanol tolerant lineages bear noteworthy similarities to mutations reported to affect substrate specificity in acrA and mexB, a Pseudomonas aerogenosa structural homolog to acrB (Figure 4B) . Mutations N154T and R59S of AcrA are spatially proximal to D111N and V244M AcrA mutations reported to affect substrate specificity of AcrA-MexB (Figure 4). Mutation V773 of AcrB is in the vicinity of the TolC docking region of MexB/AcrB, where mutation A802V of MexB is known to affect substrate specificity (Figure 4). Mutation P988L of AcrB is located in a turn between transmembrane α-helices; several MexB mutations associated with changes in substrate specificity (T329I, T557I, and T489I) also occur in turns between transmembrane α-helices (Figure 4).
marC mutations were detected in all endpoint populations, providing strong evidence of parallel adaptation at this locus (Table 2). All detected marC mutations were transposon (IS1 or IS5) insertions, with the exception of an in-frame six bp deletion in G2 (Table 2). Transpositions occurred at positions 1625925 and 1626081/1626084, suggesting that these sites are insertion hotspots. Transpositions into marC likely cause loss-of-function from disruption, and could also affect expression of the divergently transcribed marRAB operon. Functional effects of the marC six bp deletion in G2 are not immediately obvious; this mutation results in deletion of two residues (V13 and V14) from a transmembrane helix.
Mutations in mdh were also common in the evolved populations, with mutations detected in X2, X3, and G3 at approximately 100% allele frequency (8/8 clones) (Table 2). All mdh mutations were insertions or deletions resulting in frameshifts. Since substantial numbers of amino acid residues are affected in each case, these mutations are likely to cause loss-of-function of mdh. hfq mutations were less common in the endpoint populations, with mutations detected in X1 and G3 only. The X1 Hfq mutation I24M is located in the 3'-proximal purine nucleotide selectivity pocket (R-site) . The R-site is involved in binding polyA RNA, but possible functional effects of the I24M mutation are not immediately obvious . In G3, the ribosome binding site of hfq is partially deleted, potentially leading to lower intracellular Hfq protein levels through reduced translation initiation rate of hfq mRNA.
Genotypic evolutionary dynamics
We investigated the dynamics of genotypic adaptation in the G3 and X3 lineages by phenotyping and genotyping population samples from intermediate generations (Figure 5, Additional file 3). Phenotyping was done by assessing growth rate at various isobutanol concentrations, while intermediate generation genotyping was conducted by screening whole-population samples for mutations identified in sequenced clones, using Sanger sequencing of PCR amplified loci of interest or allele specific PCR. Due to the large number of mutations in the G3 end population clones, we screened only for those mutations identified in G3.266.7 and the acrB and gatZ loci (Figure 5A). All mutations detected in X3.5 were screened in the intermediate generations (Figure 5B).
The phenotype/genotype trajectories reveal that genotypic adaptations in each lineage had pleiotropic effects across different isobutanol concentrations. In both the X3 and G3 lineages, the first mutations acquired (marC/miaA-hfq in G3 and marC/gatC/hrpA/yfgO in X3) appear to drastically increase growth rates at intermediate isobutanol concentrations (1% and 0.75% w/v for G3 and X3, respectively), while having neutral or negative effects at 0% isobutanol (Figure 5). The initial marC/miaA-hfq mutations fixed in the G3 lineage appear to have a slightly negative effect on growth rate at 0% isobutanol (Figure 5A). Subsequent mutations in the G3 lineage (rph, mdh, groL, glnE, gltD, and gatZ) appear to monotonically increase the growth rate at 1% (w/v) isobutanol while gradually restoring growth rate at 0% isobutanol (Figure 5A). In the G3 lineage, the 0% and 1% (w/v) isobutanol growth rate trajectories appear to plateau after about 260 generations, while growth rate at 2% (w/v) isobutanol increases to the endpoint population (Figure 5A). In contrast, the growth rate trajectories at 0% and 0.75% (w/v) isobutanol in X3 increase to the end of the evolution, while the growth rate at 1.5% (w/v) isobutanol is relatively constant after generation 266 (Figure 5B). Interestingly, in X3 there was a period during the evolution between generations 150 and 266 where growth rate changes at 0% and 0.75% (w/v) isobutanol were flat, while there was a rapid increase in the growth rate at 1.5% (w/v) isobutanol. The growth rate increase in 1.5% (w/v) isobutanol is correlated with an mdtJ::IS5::tqsA mutation appearing at generation 180 and a mdh/deaD/plsX mutation cluster appearing in generation 266.
DNA microarray study of gene expression changes in G3.2
To gain insights into potential regulatory adaptations to isobutanol stress, we performed a gene expression study with G3.2, a highly isobutanol tolerant sequenced clone. We examined gene expression in G3.2 and the parent E. coli EcNR1 (WT) in 0% and 0.5% (w/v) isobutanol glucose minimal medium. For each strain/culture condition (G3.2/0% isobutanol, G3.2/0.5% isobutanol, WT/0% isobutanol, WT/0.5% isobutanol), three biological replicates were employed. Cultures were inoculated in media containing respective amounts of isobutanol, grown to mid log phase and harvested for transcriptome measurement. RNA samples were labelled and hybridized to a custom E. coli microarray as described in the Materials and methods section. A total of 4280 genes were included on the microarrays. After a pre-processing procedure that included background adjustment and normalization, 4235 genes with acceptable signals were subject to further analysis. Two filters were first employed to select genes with notable changes across the conditions, which resulted in a list of 2026 genes. Two-sample student's t-test was then conducted to determine statistically significant differences in gene expression. The full set of microarray results is included in Additional file 4. As illustrated in Figure 6A, 326 and 381 genes were differentially regulated by isobutanol stress in WT and G3.2 respectively. Differential transcriptional response between WT and G3.2 to isobutanol stress was observed for 223 genes, with the most significantly perturbed genes (ranked by p-value) shown in Figure 6B (see Additional file 4 for full results). Real time quantitative reverse transcription polymerase chain reaction (qRT-PCR) was used to validate two genes with large expression changes (gadA and fimI) and two genes with subtle expression changes (fabA and rfaJ) (Additional file 5). Target expression levels were determined by fitting a MAK2 model to qRT-PCR data, and expression was normalized to housekeeping gene rpoD, which was found to be invariant across all strains/conditions in our microarray data set and has been used in other studies to normalize gene expression data in gram negative bacteria [31, 32]. Expression levels measured by qRT-PCR correlated well with microarray data (Additional file 5).
BiNGO was used to assess any overrepresented Gene Ontology (GO) in the full set of genes with differential transcriptional response, using p = 0.05 as a cutoff for significance. Overrepresented ontologies included transition metal ion transport, amine transport, amino acid metabolic processes, glutamine family amino acid metabolic processes, chemical homeostasis, and various cell envelope related components and processes (including flagella and fimbriae, polysaccharide biosynthesis, and lipid metabolism); a full list of overrepresented gene ontologies and related genes is available in Additional file 4. We further investigated changes in regulation by examining transcription factors known to control genes differentially regulated between WT and G3.2. Acid fitness island genes (gadA, gadE, gadB, gadC, slp, hdeD, yhiD, hdeB, and hdeA), regulated by GadE, GadX, and GadW, are strongly repressed at both 0% and 0.5% (w/v) isobutanol in G3.2 (Figure 6B and 6C, Additional file 4). Fimbrial biogenesis genes (fimF, fimH, fimA, fimI, fimC), regulated by IHF, Lrp, and HNS, are strongly repressed in G3.2 by isobutanol; genes associated with iron acquisition (entA, entC, fepA, and cirA), regulated by Fur and CRP, are found to be repressed by isobutanol in G3.2 as well (Figure 6B and 6C, Additional file 4).
To dissect the apparently complex regulatory changes evolved in G3.2, we applied Network Component Analysis (NCA) to the microarray data to identify transcription factors with significant activity changes in G3.2 compared to WT (Figure 6D, Additional file 4). Based on previous study of isobutanol response network in E. coli and preliminary examination of our microarray data, we selected 16 transcription factors (TFs) that are potentially involved in isobutanol tolerance (ArcA, PdhR, Fnr, Fur, FlhDC, OmpR, CRP, GadE, MarA, Nac, LexA, PurR, Fis, IHF, PhoB and PhoP) for this analysis. Due to limited data (i.e. four strain/isobutanol conditions), we used a subset of four TFs in each NCA analysis and repeated the analysis for different combinations of TFs. Only TFs with consistent and significant predicted activity changes across different combinations of TFs and different replicates were retained for further analysis. GadE, PhoP, FlhDC, and MarA were subsequently found to be the most significantly perturbed TFs in G3.2 compared to WT (Figure 6D).
NCA reveals constitutively reduced activity in G3.2 of GadE, a regulator of the acid fitness island genes, and PhoP, a regulator of genes involved in Mg2+ homeostasis, resistance to antimicrobial peptides, acid resistance (including acid fitness island genes), and LPS modification (Figure 6D). FlhDC, a master regulator of flagellum biosynthesis, has increased activity in G3.2 and is not repressed by isobutanol, as in WT (Figure 6D). MarA, which regulates genes associated with response to oxidative stress, organic solvents, and heavy metals, shows increased activity at 0% isobutanol in G3.2 relative to WT, and reduced upregulation in response to isobutanol. In a previous study of the isobutanol response network in E. coli, it was concluded that activities of ArcA, PhoB, and Fur were significantly increased by isobutanol stress due to isobutanol induced quinone/quinol malfunction. We performed NCA for various combinations of ArcA, PhoB, and Fur with FlhDC, GadE, MarA, and PhoP to determine whether these results are recapitulated in our study. We found that for most tested TF combinations, ArcA, PhoB, and Fur activities are increased by isobutanol in WT EcNR1, consistent with previous results (Additional file 4 and ). Responses of ArcA, PhoB, and Fur in G3.2 differ from WT, suggesting that these transcriptional responses to isobutanol stress may have changed during evolution (Additional file 4). Especially notable is the differential response of Fur to isobutanol, with upregulation in WT versus downregulation in G3.2 observed for many tested TF combinations (Additional file 4). Many of the top differentially expressed genes identified in our microarray study are regulated by IHF, HNS, Fis, and CRP (which were incidentally identified as being significantly perturbed by isobutanol in ), suggesting that these TFs may also be involved in the differential transcriptional response between WT and G3.2 (Figure 6C).
Integrated examination of genotype and microarray expression data yields insights into the genetic basis of gene expression and transcription factor activity patterns in G3.2. One of the first mutations fixed in the G3 lineage is miaA-hfq 4407505 -7:AGGAAAA, a partial ribosome binding site deletion that is likely to reduce hfq mRNA translation. hfq is a global regulator that functions by mediating binding between a variety of sRNAs and their target mRNAs, which can alter target protein levels via effects on translation initiation or mRNA degradation . hfq is required for translation of rpoS (σ38) mRNA, the master transcriptional regulator for general stress response; thus G3.2 is expected to have lower RpoS activity . Previous work indicates that in minimal medium, flhDC is strongly repressed by RpoS, while gadE is strongly upregulated by RpoS; the activity changes observed for these transcription factors are consistent with reduced RpoS activity in G3.2 . Many other gene expression changes in G3.2 are also consistent with reduced RpoS activity (see Additional file 4). In addition to rpoS, hfq regulates numerous other genes involved in a variety of cellular processes, however since hfq regulation is post-transcriptional, many of these effects cannot be captured in a DNA microarray study . Besides possible changes in post-transcriptional regulation, microarray data indicates that rpoS is differentially regulated at the transcriptional level in G3.2 compared to WT. rpoS is upregulated in WT by isobutanol stress, consistent with previous gene expression studies  (Additional file 4). In contrast, in G3.2 rpoS expression appears to be slightly repressed by isobutanol; furthermore the basal expression level of rpoS in G3.2 is lower compared to WT, providing additional evidence of reduced RpoS activity in G3.2.
NCA analysis revealed constitutively reduced activity of the PhoP and GadE transcriptional regulators. PhoP is part of a Mg2+ responsive two-component signal transduction system, with sensor kinase PhoQ phosphorlyating (and thus activating) PhoP in response to low Mg2+ levels [35, 36]. Interestingly, G3.2 has a phoQ 1197581 A→G mutation, causing L209P in transmembrane region 2 in the PhoQ protein, which may lead to reduced activity of the PhoPQ system. Transcriptional changes caused by phoPQ perturbation are potentially adaptive, since PhoP is involved in stress response and regulates genes related to Mg2+ homeostasis, resistance to antimicrobial peptides, acid resistance, and LPS modification [35, 36]. In a previous NCA study , GadE activity was found to be strongly repressed by isobutanol. This finding was recapitulated in our NCA results for WT, while in G3.2 GadE is constitutively repressed (Figure 6D). The evolution of constitutive GadE repression in G3.2 hints that the GadE regulon (comprised of the major acid resistance genes) may be maladaptive to isobutanol stress. There is substantial overlap between the PhoP, GadE, Hfq, and RpoS regulons, pointing towards possible co-evolution between these different regulators.
Investigating phenotypic and functional effects of mutations
Previous investigations have identified the cell envelope as a primary target of solvent toxicity. G3.2 contains mutations in numerous genes and regulators associated with the cell envelope, including secA and lepB (components of the Sec apparatus, which translocates periplasmic and membrane targeted proteins from the cytosol), hfq (involved in sRNA mediated regulation of many membrane proteins), fepE and yjgQ (involved in LPS biosynthesis), and phoPQ (regulator of various LPS modification genes). Additionally, the DNA microarray study revealed that many genes related to cell envelope components and processes were differentially expressed in G3.2. We investigated possible cell envelope adaptations by profiling cellular fatty acid composition and cell envelope proteins in the parent E. coli EcNR1 strain and G3.2 during growth at 0.5% isobutanol (Figure 7A and 7B). Cellular fatty acid composition was determined using gas chromatography-flame ionization detector (GC-FID) quantification, and was found to differ considerably between G3.2 and WT EcNR1 (Figure 7A). The cyclopropane fatty acid fraction is significantly reduced in G3.2, probably as a result of downregulation of cfa (cyclopropane fatty acyl phospholipid synthase) in this strain (Figure 7A; see Additional file 4 for cfa expression data from the DNA microarray study). Additionally, the overall unsaturated:saturated fatty acid ratio is increased in G3.2 (Figure 7A), due mainly to an increase in the proportion of C16:1 and C18:1 fatty acids relative to C16:0 (data not shown).
To determine cell envelope protein profiles, cell envelopes were isolated from 5 × 109 cells by sonication and differential centrifugation and then analyzed with sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) (Figure 7B). SDS-PAGE analysis reveals an overall increase in cell envelope proteins (on a per cell basis) in G3.2 compared to WT. (Figure 7B). To examine changes in relative protein abundance between G3.2 and WT, protein bands were quantified by densitometry analysis (using ImageJ software) and normalized to the sum of intensities of the major protein bands. The 72 kDa, 55 kDa, and OmpC/OmpF bands were found to be notably upregulated in G3.2 relative to WT, with relative increases of 1.2, 2.2, and 1.3 fold, respectively (Figure 7B). Upregulation of OmpC/OmpF is consistent with DNA microarray results, which show upregulation of ompF in G3.2 (Additional file 4).
In addition to characterizing possible cell envelope adaptations in G3.2, we conducted detailed investigations of phenotypic and functional effects of key mutations identified in isobutanol tolerant clones. Selected mutations were reconstructed in E. coli EcHW24 (EcNR1 ΔmutS) singly and in various combinations using ssDNA mediated recombination . We focused on characterizing parallel genotypic adaptations, including marC, acrAB, mdh, and rph mutations identified in G3.2 and X3.5, as well as the first four mutations to appear in the G3 lineage, marC, miaA-hfq, rph, mdh, and groL, which are associated with monotonically increasing isobutanol tolerance (Figure 4A). marC, acrAB, mdh, and rph single mutants were constructed to study the phenotypic and functional effects of these mutations in isolation, while marC, miaA-hfq, rph, mdh, and groL mutations were constructed singly and in various combinations to study fitness benefits and investigate possible epistatic interactions among these mutations. Phenotypic effects were investigated by measuring growth of mutants in isobutanol spiked minimal medium, and functional assays were performed for acrAB and mdh. The parent E. coli EcNR1 and E. coli EcNR1 single gene knockouts (ΔacrA::kan, ΔacrB::kan, Δmdh::kan) were employed as controls in phenotype and functional assays.
marC mutations were detected in every evolution endpoint population. All detected marC mutations were transposon (IS1 or IS5) insertions, with the exception of an in-frame six bp deletion in G2. marC transposon insertions could not be produced with ssDNA mutagenesis, so we instead approximated the effect of transposon insertions by knocking out marC, reasoning that this could mimic effects of gene disruption caused by transposon insertion; additionally, deletion of marC was found to improve isobutanol tolerance in an independent study (James C. Liao, UCLA personal communications). Consistent with our expectations, ΔmarC::kan was found to significantly improve maximum specific growth rates and final densities in 0.5% (w/v) isobutanol minimal medium relative to the parent E. coli EcNR1 (Table 3). Growth rate improvement of ΔmarC::kan was higher in xylose medium (39 ± 2% above WT growth rate) compared to glucose medium (20 ± 5% above WT growth rate) at 0.5% (w/v) isobutanol (Table 3). In contrast, ΔmarC::kan improved final cell densities more in glucose medium compared to xylose medium (40 ± 10% vs. 7.6 ± 0.5% improvement over WT; Table 3). ΔmarC::kan had a slight negative effect on maximum specific growth rate and final cell densities at 0% (w/v) isobutanol in both xylose and glucose media (Table 3).
acrAB mutations were identified in five out of six independent evolved populations, suggesting that mutations at this locus are likely to have positive adaptive effects. Consistent with this expectation, acrA 483735 +1:A (identified in X3.5) and acrB 480665 G→A (identified in the G3 lineage) dramatically increased maximum specific growth rates and final cell densities in 0.5% (w/v) isobutanol minimal medium relative to the parent E. coli EcNR1, while having more subtle effects on growth in 0% isobutanol (Table 3). ΔacrA::kan and ΔacrB::kan produced fitness benefits of similar or greater magnitude, implying that loss-of-function of acrAB is associated with improved isobutanol tolerance (Table 3). This result is surprising given that the AcrAB-TolC efflux pump is an important mechanism of tolerance to other organic solvents and antibiotics. AcrAB-TolC efflux pump activity was measured via ethidium bromide (EtBr) accumulation in reconstructed single mutants and clonal isolates harbouring acrAB mutations from evolution end populations . Since AcrAB-TolC is the primary efflux pump for EtBr, mutations altering AcrAB-TolC activity or substrate specificity would be expected to affect the accumulation of intracellular EtBr . Increased EtBr accumulation (consistent with reduced AcrAB-TolC activity) was observed in all examined end population clonal isolates harbouring acrAB mutations (full data set in Additional file 6). The acrA 483735 +1:A single mutant had an EtBr accumulation profile similar to ΔacrA::kan and X3.5. In contrast, the EtBr accumulation profile in G3.2 was similar to ΔacrB::kan, but acrB 480665 G→A (identified in the G3.2) showed only modest changes in EtBR accumulation relative to the parent strain, implying that G3.2 may have additional mutations affecting efflux pump activity.
The AcrAB-TolC multidrug efflux pump has been well characterized in its role for antibiotic and solvent tolerance, but a recent study suggests that AcrAB-TolC may also function as an exporter for a hitherto unidentified quorum sensing signal (QSS) . There is strong evidence that the QSS exported by AcrAB-TolC is associated with upregulation of rpoS transcription; ΔacrAB mutants have reduced rpoS expression and altered temporal patterns of expression . Our gene expression study of G3.2 provides evidence of reduced RpoS activity in this strain. Interestingly, two evolved populations, X1 and G3, were found to have mutations in hfq, which is required for translation of rpoS mRNA, suggesting that RpoS modulation might be a common adaptive effect of these different mutations. We assayed RpoS activity via iodine staining in the parent E. coli EcNR1 strain, each evolution endpoint population (G1, G2, G3, X1, X2, X3), a ΔacrA::kan mutant, and a constructed single mutant containing the miaA-hfq mutation found in the G3 lineage (miaA-hfq 4407505 -7:AGGAAAA). RpoS positively regulates glycogen biosynthesis, which can be measured by staining cells with iodine - cells with higher glycogen levels stain darker . While this assay is an indirect measure of RpoS activity and is subject to many confounding factors (such as other regulation of glycogen biosynthesis), it is commonly used in literature and has been demonstrated to be well correlated with RpoS activity .
Iodine staining results are show in Figure 8A. Single mutant miaA-hfq 4407505 -7:AGGAAAA (hfq* in Figure 8A) stains lighter than the parent E. coli EcNR1 strain (WT in Figure 8A), consistent with the expected reduction of Hfq and RpoS activity in this mutant. Both of the end populations harbouring hfq mutations, X1 and G3, stain much lighter than WT suggesting reduced Hfq activity and subsequently RpoS levels in both end populations (Figure 8A). G1 and X2 also show significantly lighter staining than WT, suggesting reduced RpoS activity in these strains as well (Figure 8A). Staining in X3 is only slightly lighter than WT, while G2 and ΔacrA::kan (unexpectedly) stain very similarly to WT. Curiously, the association between ΔacrAB and reduced RpoS reported in the literature was not evidenced in our iodine staining assay. We suspect that this discrepancy may be due to differences in assay techniques. Previous studies of RpoS activity of ΔacrAB mutants were done with liquid cultures, with RpoS activity assayed by real-time PCR or Western blotting, while our assay was done on solid medium using iodine staining to measure intracellular glycogen levels, which are directly controlled by RpoS . Concentrations of the QSS exported by AcrAB-TolC are likely to vary dramatically between liquid and solid cultures due to cell density differences, and could thus confound assay results.
As a follow up to the I2 staining assay, we directly checked RpoS expression by Western blot analysis of RpoS in the parent EcNR1 strain (WT), G3.2, and single mutant miaA-hfq 4407505 -7:AGGAAAA (hfq*) grown with and without 0.5% (w/v) isobutanol (Figure 8B). RpoS Western blot analysis was repeated several times to verify results; Figure 8B shows a representative Western blot. RpoS expression is evident in the parent EcNR1 strain (WT) at both 0% and 0.5% (w/v) isobutanol, while RpoS expression was not detected in G3.2 under either condition (Figure 8B). Interestingly, in hfq* RpoS is detectable at 0% isobutanol, but not at 0.5% (w/v) isobutanol. These results directly demonstrate reduced RpoS expression at 0.5% (w/v) isobutanol in G3.2 and the miaA-hfq single mutant relative to the parent EcNR1 strain, consistent with I2 staining results (Figure 8A). We attempted Western blot analysis of RpoS in other strains (including evolution endpoint populations G1, G2, G3, X1, X2, and X3; sequenced isobutanol tolerant clone X3.5; and ΔacrA::kan single mutant); however, due to inconsistent outcomes between experiments, we are unable to draw conclusions about RpoS expression levels in these other strains (results not shown). To ascertain whether reduced RpoS activity is indeed adaptive to isobutanol stress, we examined the isobutanol tolerance phenotype of a ΔrpoS::kan mutant (Figure 8C). ΔrpoS::kan caused a growth defect at 0% and 0.5% (w/v) isobutanol relative to the WT strain, while at 1% (w/v) isobutanol the relative fitness of ΔrpoS::kan is slightly higher than WT; to facilitate comparison, we report normalized relative fitness (calculated by dividing relative fitness by relative fitness at 0% w/v isobutanol). These results suggest that attenuated RpoS activity may indeed be adaptive to isobutanol stress, since normalized relative fitness is increased at 0.5% and 1% (w/v) isobutanol (Figure 8C). However, complete loss-of-function of rpoS appears to incur significant costs that overshadow adaptive effects at isobutanol concentrations below 1% (w/v) (Figure 8C).
mdh mutations appear in three out of six evolution end populations, suggesting that these mutations may be adaptive. However, in 0% and 0.5% (w/v) isobutanol spiked minimal medium, we found relatively minor differences in growth between the parent E. coli EcNR1 strain, mdh 3390726 -1:C (found in G3.2) single mutant, mdh 3390936 +5:AACCT (found in X3.5) single mutant, and Δmdh::kan (Table 3). Thus mdh mutations do not appear to improve isobutanol tolerance in isolation, hinting that fitness benefits may come via epistatic interactions with other mutations. All of the mdh mutations identified in evolution end populations were indels causing frameshifts, suggesting that these mutations lead to loss-of-function. To assess functional effects of mdh mutations, NADH dependent malate dehydrogenase activity was measured in crude cell lysates of G3.2, mdh 3390726 -1:C single mutant, X3.5, mdh 3390936 +5:AACCT single mutant, Δmdh::kan, and the parent E. coli EcNR1. NADH dependent malate dehydrogenase activity was not detectable in G3.2, mdh 3390726 -1:C single mutant, X3.5, mdh 3390936 +5:AACCT single mutant, or Δmdh::kan, while assay of the parent E. coli EcNR1 yielded enzyme activity of 3.8 ± 0.2 U/mg-wet-cells, consistent with our expectation that 3390726 -1:C and 3390936 +5:AACCT lead to loss-of-function of mdh.
Restorative mutations are commonly observed in rph-pyrE during experimental evolution studies with E. coli K12 MG1655 , and indeed all sequenced clonal isolates from our evolution end populations had rph mutations. We investigated the adaptive benefits of the rph 3823220 +4:GTCG mutation acquired in the G3 lineage. This mutation was found to substantially improve maximum specific growth rate in both 0% and 0.5% (w/v) isobutanol spiked glucose minimal medium, consistent with the notion that rph mutations are a general adaptation to growth on minimal media (Table 3).
Genotypic adaptation to isobutanol stress is complex and involves diverse genetic loci, as revealed in our genome resequencing results. The apparent multigenic nature of isobutanol tolerance suggests that epistasis, interactions between different genes, is probably an important factor in many of the evolved genetic adaptations. To study fitness benefits and investigate possible epistasis, the first five mutations fixed in the G3 lineage (Figure 5A), marC, miaA-hfq, rph, mdh, and groL, were reconstructed singly and in various combinations in E. coli EcHW24, using multiplex recursive ssDNA mediated mutagenesis . As explained above, the marC::IS1 mutation could not be created using ssDNA recombination, so instead we knocked out marC (marC::kan) to approximate gene disruption effects caused by IS1 insertion. The resulting mutant set was phenotyped by measuring the maximum specific growth rate in 0%, 0.5%, and 1% (w/v) isobutanol glucose minimal media; results are presented as relative fitness, defined as mutant maximum specific growth rate divided by maximum specific growth rate of the parent E. coli EcHW24. Epitasis is assumed to follow a simple multiplicative fitness model , where w = relative fitness of a particular mutation combination, ε = total epistatic interaction parameter, and wi = relative fitness of single mutants; log epistasis is calculated as (Figure 9) .
As would be expected, there is a general trend of improved relative fitness with increasing numbers of mutations (Figure 9; miaA-hfq abbreviated hfq and marC::kan abbreviated marC). Relative fitness improvements at 1% (w/v) isobutanol are the most dramatic, with the miaA-hfq/rph/mdh/groL and marC/rph/mdh/groL quadruple mutants having a 3.8 fold increase in growth rate compared to E. coli EcHW24; fitness changes at 0% and 0.5% (w/v) isobutanol are much smaller, and appear to plateau with introduction of an rph mutation (Figure 9). Individually, the marC, miaA-hfq, rph, mdh, and groL mutations have relatively modest effects. mdh and groL single mutants have fitness essentially identical to the parent E. coli EcHW24 at all tested isobutanol concentrations (Figure 9). The rph single mutant has improved relative fitness at 0% and 0.5% isobutanol, while the marC and miaA-hfq single mutants have improved relative fitness at 0.5% and 1% isobutanol (Figure 9). Notable improvements in relative fitness in 1% (w/v) isobutanol were observed for some double mutants, in particular, miaA-hfq/rph, miaA-hfq/mdh, marC/mdh, and miaA-hfq/groL (Figure 9). We suspect that there may be positive epistasis between miaA-hfq and each of mdh, rph, and groL; however due to limited growth rate measurement precision, we can assert statistically significant epistasis only for miaA-hfq and mdh (Figure 9). Likewise, positive epistasis between marC and each of mdh and groL seems plausible (Figure 9), but cannot be ascertained due to limited measurement precision. Interestingly, marC and miaA-hfq demonstrate significant negative epistasis at 1% isobutanol (Figure 9).
Many higher order mutation combinations show substantial relative fitness improvements at 1% (w/v) isobutanol (Figure 9). The quadruple mutants miaA-hfq/rph/mdh/groL and marC/rph/mdh/groL have the greatest relative fitness, with a 3.8 fold improvement in growth rate, followed by the marC/miaA-hfq/rph/mdh/groL pentuple mutant and hfq/rph/mdh triple mutant, each having a 3.3 fold improvement in growth rate (Figure 9). Results for the reconstructed marC/hfq/rph/mdh quadruple mutant and full pentuple mutant are approximately consistent with the evolution trajectory (Figure 5A). Epistasis analysis reveals significant positive epistatic interactions for miaA-hfq/rph/mdh, rph/mdh/groL, miaA-hfq/rph/mdh/groL, and marC/rph/mdh/groL (Figure 9). Comparison of fitness effects of rph/mdh/groL vs. miaA-hfq/rph/mdh/groL and marC/rph/mdh/groL provides compelling evidence that an miaA-hfq or marC genetic background exhibits positive epistasis with rph/mdh/groL (Figure 9).