Experimental evolution can compensate for defects caused by the deletions of chemotaxis genes

Experimental evolution of E. coli mutant strains was performed under selection for increased spreading on tryptone broth soft-agar (TBSA) plates for 30 cycles of up to 16 hours each (S1A Fig), and with up to four independently evolved lines. All these strains were derived from E. coli strain RP437 that is commonly used as the wildtype for studies of chemotaxis [25]. We relied on the natural mutation rate of E. coli cells, which is known to be sufficiently high to enable evolutionary tuning of motility under strong selection for chemotaxis in TBSA [22].

Consistent with numerous previous studies, none of these strains deleted for individual chemotaxis genes initially exhibited spreading in soft agar (Figs 1B-D, and S1B). However, the spreading of the evolved ∆cheR, ∆cheB, and ∆cheZ strains significantly improved compared to the original deletion strains in all evolved lines (Fig 1B). These three strains lack the auxiliary components of the signaling pathway, either the adaptation enzymes or the phosphatase (Fig 1A). The extent of the improvement varied between gene deletions and lines, with the largest improvement being observed for ∆cheR lines where spreading reached ~50% of the non-evolved wildtype (wt) (Fig 1C). Notably, a previous study has shown that the spreading of the RP437 wildtype could itself be enhanced up to 50% by the experimental evolution under similar conditions [22], meaning that the evolved ∆cheR lines in our experiments could achieve ~30% spreading of the evolved wildtype.

In contrast, the absence of universally conserved core components of the chemotaxis pathway that are required for sensing and signal transduction to the motor, including CheA, CheY, CheW, or all chemotaxis receptors (MCPs), could not be compensated for by such short-term evolution (Figs 1D and S1B). Furthermore, no improvement in the spreading of a double ∆(cheR cheB) deletion strain under selection was observed (Fig 1E). This was surprising, since the ∆(cheR cheB) strain prior to evolution was spreading slightly better than the individual cheR or cheB deletion strains, likely because of its intermediate tumbling bias [14]. As the ∆cheR strain showed the largest enhancement of spreading, we evolved four additional ∆cheR lines (S2 Fig).

Evolved strains exhibit compensatory changes in motility

We next investigated changes in the motility phenotypes of the evolved E. coli strains, by tracking cell swimming in liquid. Consistent with the known importance of intermediate tumbling frequency for spreading in soft agar [14,26], the evolved strains exhibited compensation for the defects in tumbling that were present in the original deletion strains. In nearly all cases, the fraction of time that cells spent tumbling became more similar to the wildtype (Fig 2A), with all evolved ∆cheR lines showing increased tumbling and all evolved ∆cheZ lines showing decreased tumbling compared to the respective non-evolved deletion strains. Cell tumbling in liquid correlated well with the increased spreading in TBSA (Fig 2B), consistent with a recent report showing that the tumbling bias of wildtype E. coli cells is approximately optimal for spreading in soft agar [26]. However, the evolutionary adjustment of tumbling was clearly not the sole determinant of the improved spreading, as strains with similar tumbling could spread very differently. This is exemplified by ∆cheB lines, where spreading of several lines has improved without changes in tumbling.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Acquired changes in motility phenotypes and flagellar gene expression in the evolved lines.

(A) Tumbling bias, defined as the fraction of time spent tumbling, for the wildtype, parental non-evolved strains (R0, B0 and Z0), and evolved strains (R1-R4, B1-B4, Z1-Z4), measured in three independent replicates. (B) Size of the spreading rings in TBSA, normalized to that of the wildtype strain (data from Fig 1C), plotted as a function of tumbling bias for individual strains. The dotted line indicates the tumbling bias for the wildtype. Inset: Spreading in TBSA as a function of the absolute deviation of the tumbling bias from that of the wildtype strain. The relationship between the two variables was evaluated using Spearman’s rank correlation, yielding a coefficient of –0.51. (C) Run speed between two consecutive tumbles, measured for all strains in three independent replicates. Motility phenotypes were assessed using cell tracking (see Materials and Methods). (D) Activity of transcriptional fliC promoter (PfliC) reporter, measured as fluorescence of green fluorescent protein (GFP) using flow cytometry in three independent replicates. Inset: Run speed (data from C) plotted as a function of the fliC promoter activity. Colors correspond to protein colors in Fig 1A. Significance analysis was done in comparison to the respective non-evolved deletion strains. Error bars indicate standard errors of the mean. P values were calculated using two-tailed Student t-test (ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001).

https://doi.org/10.1371/journal.pgen.1011784.g002

Furthermore, nearly all of the evolved strains exhibited an increased swimming velocity (Fig 2C). Similar increase in velocity was previously observed during the evolution of wildtype cells for the spreading in TBSA, and it was shown to be the consequence of the elevated expression of the flagellin gene fliC and other flagellar genes [22]. Therefore, we measured activity of the transcriptional reporter of fliC promoter (PfliC) [22]. This activity was indeed significantly higher in most of the evolved strains compared to their parental strains (Fig 2D), consistent with our hypothesis that the increased swimming velocity of the evolved strains may be primarily due to the changes in flagellar gene expression. Cell swimming velocity was previously shown to initially increase with the flagellar regulon activity above that in the wildtype RP437 but subsequently saturate at approximately twofold higher expression levels [22,27], which could be observed in our data as well (Fig 2D Inset).

Since spreading in soft agar requires not only motility and chemotaxis but also cell growth, we further quantified growth of the evolved strains. We observed that their growth rate generally decreased compared to the parental strains (S3 Fig). This observation could be at least partly explained by the known negative tradeoff between expression of flagellar genes and E. coli growth [22–24,27–29], and previous experimental evolution of the wildtype in TBSA indeed led to the elevated motility at the cost of slower growth [22]. Nevertheless, there was no simple correspondence between fliC promoter activity and growth of individual evolved strains, suggesting that other factors besides flagellar gene expression contribute to the reduction of their growth rate.

Deletion-specific sets of compensatory mutations are observed in the evolved strains

Whole-genome sequencing revealed multiple mutations in the evolved ∆cheR, ∆cheB and ∆cheZ lines (S1, S2 and S3 Tables and S1 Data). These mutations exhibited clear gene-deletion specific patterns and primarily affected chemotaxis or motility genes. The most prominent set of mutations in the evolved ∆cheR lines mapped to the tsr gene that encodes the serine chemoreceptor (Fig 3A and S1 Table). Mutations in tsr were observed in seven out of eight lines. One line (R7) carried instead a mutation in tar, which encodes the aspartate receptor (Fig 3B), and two lines (R3 and R8) carried mutations in both tsr and tar (S1 Table). The corresponding amino acid substitutions affected different receptor domains, including the ligand-binding domain, transmembrane helix, HAMP domain, methylation region, and in the signaling domain, which is similar to substitutions observed in the previous report [17]. We hypothesized that these mutations may promote the active state of Tsr or Tar. This could increase the pathway activity in the evolved ∆cheR lines relative to the parental ∆cheR strain, thus contributing to the restoration of their tumbling (Fig 2A). M249 and L263 residues of Tsr were previously shown to be critical for the helical packing of the HAMP bundle and for signal transduction [30], and T305 is located adjacent to one of the methylation sites that control receptor activity [31]. Amino acids substitutions at these residues are thus likely to affect the activity state of Tsr, and other substitutions in Tsr and in Tar might be similarly activatory. Notably, mutations in tsr may have been preferentially selected over those in tar because the Tsr ligand serine is consumed faster than other amino acids [32], and the gradient of serine is thus the first gradient followed by spreading chemotactic E. coli cells on TBSA plates [33].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Evolutionary selected amino acid substitutions in chemotaxis proteins.

(A-E) Substitutions identified in the evolved R strains (red), B strains (dark blue) and Z strains (black) in (A) Tsr, (B) Tar, (C) CheB, (D) CheZ, and (E) CheA, mapped on the respective protein structure. For chemoreceptors, functional domains are labeled. For CheB, the receiver and methylesterase domains are labeled and residues in the phosphorylation and catalytic pockets are marked in orange. (F) Size of the spreading rings in TBSA of the parental ∆cheR (R0) strain carrying either individual mutations that were identified in the R1 strain (indicated by asterisks; see S1 Table) or their combinations, compared to spreading of R0 and R1 strains and of R1 strain carrying deletion of cheB. Values were measured in three independent replicates and normalized to the spreading ring size of the wildtype strain. Error bars indicate standard errors of the mean. Significance analysis was done in comparison to R0. P values were calculated using two-tailed Student t-test (ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001).

https://doi.org/10.1371/journal.pgen.1011784.g003

Most of the evolved ∆cheR lines also had mutations in the cheB gene (Fig 3C and S1 Table). CheB consists of a regulatory CheY-like receiver domain, which is phosphorylated by CheA, and the enzymatic methylesterase domain [34]. Mutations were found in both domains, with an apparent clustering near the catalytic pocket, the phosphorylation pocket, and the regulatory interface [35]. These mutations may lower the methylesterase activity of CheB, as previously suggested [16], which could potentially compensate for the absence of methyltransferase activity. Nevertheless, no amber or frameshift mutations were detected, indicating that certain level of CheB activity is necessary for the restored spreading of the evolved ∆cheR lines (see below). Finally, two of the ∆cheR lines had mutations in cheZ (Fig 3D). Q204L affects the C-terminal CheY-binding peptide [36] and Q64L is close to the catalytic site of CheZ [37]. Substitutions at both sites were reported to lower the phosphatase activity of CheZ [38], suggesting that they can partly offset the low kinase activity in the ∆cheR strain and thus increase the tumbling bias in the evolved lines.

Mutations in the receptor genes tsr, tar and tap were also found in all ∆cheB lines (Fig 3A and 3B, and S2 Table). Since the lack of CheB causes excessive receptor methylation and therefore hyperactivity, we would expect these mutations to counter this lack by favoring the low activity state of receptors. Additionally, one amino acid substitution affected the dimerization domain of CheA, and a short ALGD amino acid sequence was inserted in CheW (Fig 3E and S2 Table). These mutations may similarly affect the activity or stability of the ternary receptor-CheA-CheW complex, offsetting the hyperactive receptor phenotype caused by the ∆cheB deletion.

The evolved ∆cheZ lines have similarly compensated the hyperactive pathway phenotype but with a different set of mutations (S3 Table). In all lines, Tar translation was interrupted either by a stop codon mutation within the ligand-binding domain or by a frameshift in the transmembrane helix of the receptor (Fig 3B). These mutations likely lower the overall pathway activity by reducing the overall number of receptor proteins. Additionally, the disruption of tar results in more uniform composition of the chemosensory complexes that primarily consist of the remaining major receptor Tsr, which may increase sensitivity of the chemotactic response to the gradient of Tsr ligand serine [39] that is the first attractant gradient encountered by the spreading colony (see above). Each of these mutations was observed in two of the evolution lines, possibly because they were already present at low frequency before evolution, due to standing genetic variation within population. All ∆cheZ lines also had mutations in cheA, either in the P1 (phosphorylation) domain or in the P4 (kinase) domain (Fig 3E). These mutations may reduce the activity of CheA and consequently lower the phosphorylation of CheY, thus compensating for the absence of CheZ.

In addition to these mutations in the chemotaxis genes, almost all ∆cheR lines and one ∆cheB line had mutations in the genes that encode the export apparatus (fliI) and the basal body (fliF, fliM, fliN, and fliG) of the flagellar motor (S1 and S2 Tables). Similar mutations have been previously shown to upregulate the expression of flagellar genes by enhancing secretion of the negative regulator (anti-sigma factor) FlgM [22], which is consistent with the elevated activity of PfliC reporter in these lines (Fig 2D). However, amino acid substitutions in FliM, FliN and FliG might also affect cell tumbling. Some evolved lines had no mutations in the export apparatus or in the basal body genes yet showed increased flagellar gene expression, possibly due to mutations or insertions in other genes such as clpX, sspA or rpoD, which are known to affect the expression of the flagellar regulon [22,40] (S1, S2 and S3 Tables). Finally, the mutations in atp genes encoding the PMF-dependent ATP synthase may increase motility by elevating the proton motive force.

Chemotactic spreading is restored by cumulative effects of multiple mutations

We further investigated the order in which mutations were detected over the course of evolution for several ∆cheR lines (S4A, S4B, S4C and S4D Fig), which revealed that the mutations in tsr were selected first, followed by mutations in other chemotaxis and/or flagellar genes. One of the best-spreading lines, R1, was subsequently selected to evaluate the phenotypic impacts of individual mutations and their potential epistatic interactions. R1 carries mutations in the chemotaxis genes tsr, cheB and cheZ (Fig 3A, 3C and 3D), and a mutation in the flagellar export gene fliI. When these mutations were introduced individually into the ∆cheR strain, mutations in tsr or cheB significantly increased spreading on TBSA plates (Fig 3F). Indeed, these two genes are most commonly affected in the evolved ∆cheR lines. Spreading was further increased by combinations of multiple chemotaxis and fliI mutations. Interestingly, the observed order of selection for the individual mutations during evolution of the R1 line (S4A Fig) is apparently consistent with the path of the largest stepwise increase in spreading due to addition of each subsequent mutation, tsr < tsr cheZ < tsr cheZ fliI < tsr cheB cheZ fliI. Spreading of the ∆cheR tsr cheB cheZ fliI strain largely recapitulated that of the R1 line, with the residual difference being likely due to additional mutations present in the R1 line. Indeed, inactivation of sspA – which was interrupted in R1 – was previously shown to increase flagellar gene expression [22], and the introduction of this mutation led to further increase in spreading (Fig 3F). A gradual increase in spreading was similarly observed when cheB and tsr mutations from R4 and R5 lines were introduced individually into the ∆cheR strain (S4E Fig). Similar to the R1-specific mutations, the effect of tsr mutations on spreading was stronger than that of cheB mutations.

We tested the effects of R1-specific mutations on the chemotactic spreading when introduced individually in the wildtype cells (S4F Fig). Mutation in fliI led to enhanced spreading, consistent with the previous report where similar mutations were shown to increase swimming velocity [22], while all other mutations resulted in no or only modest changes in spreading in the wildtype background. This indicates that all of the mutated genes remain functional, including mutated cheB and cheZ genes that are expected to produce proteins with reduced enzymatic activities to compensate for the lack of CheR. To directly test whether the function of CheB was required for the spreading of the R1 line, we introduced the cheB deletion in the evolved strain. The R1 strain lacking cheB showed even less spreading in soft agar than the ∆cheR strain (Fig 3F), confirming that the (reduced) activity of CheB is necessary for the re-evolved spreading of ∆cheR strains.

To verify the impact of mutations in cheB and in tsr genes on the functionality of their products, we characterized the pathway response in the wildtype cells carrying the R1-specific mutations in these genes (S5 Fig). This was done using the previously established assay based on Förster (fluorescence) resonance energy transfer (FRET), which monitors the phosphorylation-dependent interaction between CheY fused to a yellow fluorescent protein (CheY-YFP) and its phosphatase CheZ fused to a cyan fluorescent protein (CheZ-CFP) [41,42]. High pathway activity leads to increased phosphorylation of CheY-YFP, which results in increased complex formation with CheZ-CFP and higher energy transfer from the excited CFP donor to the YFP acceptor fluorophore, reflected by the increased ratio between the YFP and CFP fluorescence emission. Consistent with the previous work [41], inhibition of the pathway activity in the wildtype cells by stimulation with attractant (MeAsp) leads to the adaptive pathway response, characterized by the initial decrease in FRET and subsequent recovery of the pathway activity upon prolonged stimulation (S5A Fig). This recovery is mediated by the CheR-dependent receptor methylation [41]. It is followed by a characteristic overshoot of the pathway activity upon the removal of attractant, also followed by the recovery of activity, this time due to the receptor demethylation mediated by CheB. Although the R1-specific mutation in cheB (R75C) introduced in the wildtype cells did not reduce their spreading in soft agar (S4F Fig), it had a clear impact on the pathway response (S5B Fig). The strain carrying this mutation showed larger amplitude of response to the addition of saturating concentration of attractant and weaker amplitude of response to its removal, suggesting that it has higher steady-state pathway activity. Moreover, this strain showed markedly slower adaptation to the removal of attractant, mediated by CheB. These observations are consistent with our expectation that this mutation reduces the activity of CheB. Similar elevation of the adapted pathway activity was also observed upon the introduction of the R1-specific mutation in tsr (T305M) (S5C, S5D, S5E, S5F and S5G Fig). Moreover, this latter mutation affected the dose dependence of the pathway response to the Tsr ligand serine, shifting it to higher concentrations and making it apparently steeper compared to the wildtype response (S5C, S5D and S5G Fig). In contrast, it had little effect on the dose dependence of the response to MeAsp, the ligand of Tar (S5E, S5F and S5G Fig).

Evolved strains exhibit directional spreading in chemical gradients

Although previous studies concluded that spreading in TBSA does not necessarily require chemotaxis but only an intermediate cell tumbling frequency, such pseudotaxis is much slower than the chemotaxis-driven spreading [14,43,44]. In contrast to that, spreading of some of the evolved strains, in particular of several ∆cheR lines, was comparable to that of the wildtype cells, including a characteristic ring at the edge of the spreading colony that normally indicates chemotactic behavior [43,44]. We thus tested the ability of these strains to follow chemical gradients, first using the M9 minimal medium soft agar (M9SA) gradient plates with the pre-established gradient of chemoeffector [45,46]. Three out of four ∆cheR lines showed efficient biased spreading up the gradient of serine in M9SA plates that was similar to that of the wildtype strain, again forming a characteristic sharp ring at the edge of the colony spreading up the gradient of serine (S6A and S6D Fig). Biased spreading was also observed for two out of four ∆cheZ lines (S6C and S6D Fig), although it was less pronounced than for the ∆cheR lines. The other ∆cheR and ∆cheZ lines, as well as all ∆cheB lines (S6A, S6B and S6C Fig) showed little spreading or growth on M9SA plates, so their bias could not be determined.

Although the observed spreading bias in serine gradient is indicative of chemotaxis, its interpretation is complicated by the fact that serine is metabolized. We thus tested spreading of several ∆cheR lines using M9SA plates with a gradient of α-methyl-D, L-aspartate (MeAsp), a non-metabolizable analogue of aspartate. Indeed, a reproducible and significant biased movement up the gradient of MeAsp could be observed for the R1, R4, and R5 lines, but not for R0 (ancestor) (Fig 4A and 4B). Notably, this spreading bias of these ∆cheR lines was weaker than in the serine gradient and no sharp ring at the spreading edge was visible. This difference might be due to the lower steepness of the MeAsp gradient in soft agar, as it is not metabolized by the spreading colony, or to different impacts of the selected mutations in tsr on the pathway response to serine and to MeAsp, as observed for the R1-specific mutation (S5 Fig). No spreading was observed for the R3 line that carries an amino acid substitution (A166P) in the ligand binding domain of Tar (Fig 3B and S1 Table), which might render it unable to sense MeAsp. Similarly, all ∆cheZ lines possess only a truncated version of Tar (Fig 3B and S3 Table), and therefore cannot sense MeAsp.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Biased movement of the evolved ∆cheR strains towards sources of a non-metabolizable chemoattractant.

(A) Indicated strains were tested for biased spreading on the M9 minimal medium soft-agar (M9SA) plates with a pre-established gradient of α-methyl-D, L-aspartate (MeAsp), a non-metabolizable analog of aspartate, with the concentration of 50 mM at the source (dashed line). Scale bar is 1 cm. (B) Spreading bias was quantified as the ratio between the distances from the inoculation point of the expanding colony to its edges facing up and down the gradient, and measured in three independent replicates. Error bars indicate standard errors of the mean. Significance analysis was done in comparison to 1 (no bias). P values were calculated using one-tailed Student t-test (ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001). (C) Chemotactic cell accumulation of the fluorescently labelled wildtype, ∆cheR and evolved ∆cheR strains towards 50 mM MeAsp at the source in the microfluidic device shown in S7A Fig. Examples of accumulation for the wildtype, R0, R1 strains are shown in S7B Fig. Total fluorescence intensity was quantified in the observation area (depicted in the dashed rectangle in S7A Fig) at indicated time points and normalized to the fluorescence at the initial time point. Error bars indicate standard errors of the mean.

https://doi.org/10.1371/journal.pgen.1011784.g004

To next test the ability of evolved strains to perform chemotaxis in liquid, we used a microfluidic accumulation assay where an attractant gradient is generated in the liquid medium within the test channel [45–47] (S7A Fig). The wildtype strain accumulated rapidly toward the source of MeAsp in this assay, in contrast to the fully motile but non-chemotactic R0 strain, whereas an intermediate but clear accumulation was observed for the evolved ∆cheR lines (Figs 4C and S7B). These results demonstrate that the evolved cells lacking this normally essential pathway component show biased spreading in attractant gradients in both, soft agar and liquid, suggesting that they regained the ability to perform chemotaxis, albeit less efficiently than the wildtype cells.

Chemotactic drift of the evolved cheR strain relies on greater extension of runs up the gradient without short-term adaptation

To gain an insight into the mechanism of this re-evolved chemotactic behavior, we analyzed the movement of individual swimming cells of the R1 line in a microfluidic chemotactic chamber with or without a linear gradient of MeAsp (S8A and S8B Fig). Expectedly, the R1 cells, as well as the wildtype and R0 cells, showed no biased motion, and thus no chemotactic drift, in the absence of a gradient (Fig 5A). The R0 cells also showed no bias in the gradient of MeAsp, confirming again that these cells are not capable of performing chemotaxis. In contrast, cells of the R1 line showed a clearly pronounced chemotactic drift, particularly in the gradient with 1 mM MeAsp at the source (Fig 5A). This drift could be largely recapitulated by the ∆cheR strain carrying tsr cheB cheZ fliI sspA mutations, in agreement with the measurements of spreading in TBSA (Fig 3F).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Chemotactic drift and run duration of the R1 cells in a gradient of MeAsp.

(A-B) Chemotactic drift (A) and chemotaxis bias (B) of indicated strains in the MeAsp gradients quantified with cell tracking. E. coli MG1655 wildtype strain was used as control. Error bars represent standard error of the mean, the values for each replicate are shown with separate dots. Significance analysis was done with respect to zero using two-tailed Student t-test. (C) Distributions of cell run durations for the indicated strains, either in the absence of a gradient (0-0 MeAsp, top panels) or in the presence of a gradient from zero to 1 mM MeAsp (0-1 mM MeAsp, bottom panels). Cell runs were measured using cell tracking and separated based on their direction, either up or down the gradient (S8A Fig, Materials and Methods). Mean run durations are shown in the insets. (D) Average run durations up and down measured in at least three replicated experiments. Error bars indicate standard error of the mean, the values for each replicate are shown with separate dots. To compare run durations up and down the gradient for the same MeAsp concentration, significance values were calculated using paired one-side Student t-test, while a non-paired Student t-test was used to compare run durations up (or down) in the absence and in the presence of the gradient. P value asterisks stand for: ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001.

https://doi.org/10.1371/journal.pgen.1011784.g005

Surprisingly, the drift of the R1 cells up the gradient was even slightly faster compared to the wildtype cells, which could be partly due to their higher swimming velocity. Indeed, the values of the chemotactic bias (chemotactic drift normalized by the swimming velocity) of the wildtype and R1 cells, as well as that of the cells carrying tsr cheB cheZ fliI sspA mutations, were very similar (Fig 5B). Thus, the chemotactic ability of the evolved R1 line is comparable to that of the non-evolved parental wildtype strain RP437. Nevertheless, these values remained below those observed for the experimentally evolved faster-swimming RP437 cells [22] or for the generally better motile MG1655 cells (Fig 5A and 5B) [48,49].

We next compared the distributions of the duration of cell runs up and down a gradient along the channel. All strains expectedly showed similar distributions in both direction in the absence of a gradient (Fig 5C, upper panels, and Fig 5D). A non-chemotactic (and nearly unresponsive) ∆cheR (R0) strain showed no significant difference between the duration of runs in either direction even in the presence of the gradient source. In contrast, runs of the wildtype cells (both RP437 and MG1655) became significantly elongated towards the source of the attractant and marginally shortened away from it (Fig 5C, lower panels, and Fig 5D). Notably, the average run duration of the wildtype cells in the gradient was similar to that in the absence of the attractant (S8C Fig). This agrees with the established chemotaxis strategy of E. coli, which primarily relies on the extension of cell runs up the gradient [6], above a constant average run duration that is adjusted by the adaptation system irrespectively of the current background stimulation.

The behavior of the R1 cells, and of the ∆cheR tsr cheB cheZ fliI sspA cells that carry the key R1 mutations, was distinctly different from that of the wildtype or ∆cheR (R0) strains. The R1 cells showed a noticeably shorter average run duration without stimulation, consistent with their increased tumbling (Fig 2A), and their runs became strongly extended both up and down the gradient (Fig 5C and 5D). Thus, in contrast to the parental ∆cheR strain, and similar to the wildtype, the R1 line is clearly responsive to the attractant. Moreover, both the R1 cells and ∆cheR tsr cheB cheZ fliI sspA cells showed a significantly larger elongation of runs up the gradient, which may form the basis of their chemotactic drift. But both strains are apparently unable to efficiently adjust the average cell running behavior to the background concentration of attractant (S8C Fig), consistent with the absence of CheR-mediated adaptation.

To directly test the ability of the evolved ∆cheR lines to adapt, we investigated their pathway response to attractant using the FRET assay. The wildtype cells responded to the addition and subsequent removal of different MeAsp stimuli in a characteristically dose-dependent and adaptive fashion (Fig 6A), as already discussed above (S5A Fig). No measurable FRET response could be observed for R0 cells (S9A Fig), consistent with expectedly low pathway activity in these cells. In contrast, the pathway response of R1 cells to MeAsp stimulation was well pronounced and observed in a similar concentration range as for the wildtype cells, but it showed no apparent adaptation on the time scale of minutes in our experiment (Fig 6B). Similarly non-adaptive response was observed in the R4 line (S9B Fig). Thus, the FRET assay confirmed that the evolved ∆cheR lines did not regain the ability to adapt to stimulation on the timescale relevant for the gradient sensing by swimming cells via some alternative mechanism [50].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Pathway response and slow adaptation of tumbling bias in the evolved R1 strain.

(A-B) YFP/CFP ratio, proportional to the chemotaxis pathway activity, for the populations of the wildtype RP437 (A) and the evolved R1 (B) cells expressing the FRET pair CheY-YFP/ CheZ-CFP. Cells were immobilized in a flow chamber under flow and exposed to the addition (down arrows) and subsequent removal (up arrows) of indicated concentrations of MeAsp. In both cases, the ratio was corrected for the baseline drift and normalized to its value at the beginning of the experiment. (C) Dependence of the relative kinase activity, calculated from initial changes in the YFP/CFP ratio (see S9 Fig and Materials and Methods), on MeAsp concentration for the wildtype and R1 cells. Experimental data was fitted by a Hill function. In case of R1, an additional parameter accounting for non-zero activity upon high MeAsp stimulations was added to the fitting function. Half-inhibitory concentrations (K_D) are shown in the plot. (D-E) Mean tumbling bias for the wildtype RP437 (D) or R1 (E) cells quantified with cell tracking analysis. Cells were kept in motility buffer supplemented with either only 1% glucose (Glc) or additionally with 50 µM or 500 µM MeAsp, as indicated. Motility measurements were performed in the beginning of the experiment and after 1.5 h and 3 h of incubation. Significance analysis was performed across three replicated experiments; the values for each replicate are shown with separate dots. P values were calculated in relation to the starting time point using two-tailed Student t-test (ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001).

https://doi.org/10.1371/journal.pgen.1011784.g006

Furthermore, the R1 cells have apparently higher pre-stimulation pathway activity compared to the wildtype cells, as suggested by their stronger response to attractant and weaker response to repellent (NiCl₂) (S9C, S9D and S9E Fig). This is consistent with higher tumbling bias and shorter duration of their runs in the motility buffer (Figs 2A and 5C). Moreover, the dose dependence of the R1 pathway response was moderately shifted to higher concentrations of MeAsp and, in contrast to the wildtype cells, their pathway activity was not fully inhibited by the addition of high concentrations of MeAsp (Figs 6C and S9E). These observations suggest slightly reduced sensitivity of the R1 cells to MeAsp compared to the wildtype.

The evolved ∆cheR cells have thus regained not only their responsiveness to chemoeffector stimulation but also the ability to differentially modulate their motility dependent on the direction in a gradient. This apparently occurs without short-term (seconds to minutes) adaptation of the pathway activity that forms the basis of the wildtype chemotaxis strategy. Nevertheless, we observed that the R1 cells are capable to at least partially recover their tumbling bias when exposed to MeAsp for prolonged periods of time (several hours), in contrast to the wildtype cells that adapt rapidly and subsequently maintain stable tumbling bias (Figs 6D and 6E, and S9F). Although such very slow adaptation is unlikely to directly contribute to the ability of the R1 cells to follow chemical gradients in liquid, it may be important for the maintenance of intermediate tumbling bias, and thus responsiveness to gradients, over a range of background attractant levels.

Computational simulations can partly explain the adaptation-less taxis of the evolved ∆cheR cells

To understand whether the measured properties of the pathway and swimming responses could explain the biased tactic motion of the evolved ∆cheR cells, we computationally simulated the behavior of these cells in a gradient. The cells were modeled as run-and-tumbling agents which bias their tumbling probability according to the currently experienced concentration of attractant (S1 Text). To reflect our experimental setup, we simulated cell swimming in a linear gradient of MeAsp along a 2-mm long rectangular channel with open ends connected to infinite reservoirs, which absorb and randomly respawn cells at the extremities (Fig 7A). Their concentration-dependent tumbling bias was adjusted accordingly to the experimentally measured dependence on MeAsp concentration (Figs 7B and S10A). This included the observed residual tumbling bias at even high concentration of MeAsp, which is also consistent with the FRET measurements (Fig 6C). Furthermore, we either assumed a homogeneous population of cells with the same population-averaged tumbling bias or a heterogeneous population with the experimentally measured distributions of tumbling bias (Figs 7C and S10B). Notably, the experimentally measured tumbling bias distributions of both the R1 line and of our wildtype RP437 strain were broader than that of the canonical E. coli MG1655 strain (S10C Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Simulated non-adaptive strains with R1 tumbling properties are capable of chemotaxis-like behavior.

(A) Schematics of the simulation box featuring a chemoattractant gradient, solid walls on its sides and two open ends each connected to an infinite cell reservoir (see S1 Text for details). (B) Measured population-averaged tumbling bias of the R1 cells in the presence of various concentrations of MeAsp, fitted by the homogeneous and heterogeneous models of tumbling bias response to MeAsp. (C) Experimental distributions of tumbling bias measured in the same conditions as in (B), and the corresponding inferred distributions of tumbling bias using the heterogeneous population model. Experimental data in (B,C) are from a single representative experiment. (D) Simulated average drift of the population of cells in response to the indicated gradients of the attractant for the two models and with or without the slow adaptation, as indicated. Error bars are standard errors of the mean, with N 1.5 × 10⁴ trajectories per condition.

https://doi.org/10.1371/journal.pgen.1011784.g007

In our simulations, cells with either a uniform or a heterogeneous tumbling bias were able to drift on average up the gradient (Fig 7D), with higher drift speed being observed for a steeper gradient. The simulated chemotactic drift was markedly enhanced by the heterogeneity of tumbling bias within population, where individual cells are simulated to respond sharply but at different concentrations of MeAsp [51]. This enables the non-adapting population as a whole to maintain strong response over a wider range of attractant concentrations. Our simulations thus confirm that the space-dependent modulation of tumbling bias in a gradient of attractant could alone be sufficient to elicit chemotactic drift, even without short-term adaptation at the pathway level. Nevertheless, the simulated drift remained several times smaller than the one measured experimentally, even for simulations that took population heterogeneity into account. We further tested the impact of the experimentally observed gradual recovery of the tumbling bias on the timescale of hours upon prolonged attractant stimulation. However, no significant improvement of the chemotactic drift by such slow adaptation could be observed in our simulations (Fig 7D), likely because this recovery is too slow to make a direct impact on gradient sensing. The evolved ∆cheR cells thus appear to rely on other features that are not captured by the model, such as the dynamics of flagellar motor remodeling [52], to further enhance their ability to follow chemical gradients.

Strains and growth conditions

Derivatives of E. coli strain RP437 [25], used in this study as the wildtype for chemotaxis, are listed in S4 Table. Unless specified otherwise, cultures were grown at 37°C in Luria broth (LB) with shaking at 200 rpm or on LB plates containing 1.5% (w/v) agar. For spreading ring measurements, E. coli strains were grown at 30°C for ~16 h in Tryptone broth (TB) containing 0.27% (w/v) agar (Tryptone Broth Soft Agar - TBSA). For growth rate measurements, bacterial cultures were grown in 24-well plates with transparent bottom in a plate reader (Infinite M1000 PRO, TECAN) in TB at 30°C with shaking at 180 rpm.

Experimental evolution of the chemotaxis mutants

Experimental evolution of E. coli mutant strains was conducted under selective conditions to enhance spreading on Tryptone broth soft-agar (TBSA) plates. The evolution was performed in the absence of mutagenesis and relied on the natural mutation rate of E. coli. The chemotaxis mutants were evolved through consecutive passages of sub-culturing in TBSA. The inoculum for the initial subculture was taken directly from glycerol stocks. For subsequent reinoculations, 2 µl cells were taken from the outer edge of the spread ring after overnight growth at 30°C on the TBSA plate and inoculated onto a fresh TBSA plate (S1A Fig). This process was repeated daily for 30 days, with additional samples taken for glycerol stocks. The resulting strains were sequenced as described below, and the variant calling analysis was done to identify the SNPs, insertion and deletion in the genome. The presence of these mutations in evolved strains at different stages of evolution was then analyzed using PCR amplification and Sanger sequencing. This also confirmed that the mutations were not present at detectable levels in the original populations of non-evolved strains.

SNP-reintegration/genome editing

For the introduction of the point mutations, cassettes containing the neo-ccdB were amplified from pKD45 (S4 Table) using specific primers, which include a 50-bp homology from both sides for the desired region. This cassette was introduced in the desired E. coli recipient strain containing pKD46, and strains carrying this cassette and pKD46 were then transformed with a fragment containing the desired region amplified from the evolved lines using electroporation. Transformed cells were grown in LB + 0.2% arabinose for 4–5 h at 30°C, followed by selection on M9 minimal media plates containing rhamnose as sole carbon source. Rhamnose-resistant colonies were selected and sequenced to confirm mutations.

Re-sequencing of bacterial genomes

The genomic DNA of the evolved bacterial population were isolated using the Qiagen DNeasy Blood and Tissue kit following the manufacturer’s instructions. Libraries were constructed using Nextera XT Index Kit (24 indexes, 96 samples) Illumina FC-131–1001. Sequencing was done using the Illumina HiSeq Rapid Run (2 × 150 bp paired-end run). The genomes were reassembled and mutations mapped with the DNASTAR Seqman NGen 12 software and BRESEQ pipeline, using the E. coli RP437 genome as the template. For identifications of mutations by Sanger sequencing, gene fragments of interest were amplified using PCR. This PCR fragment was subsequently purified and sent for Sanger Sequencing to Microsynth Seqlab GmbH.

Gradient plate assay

Minimal A agar (0.25% agar, 10 mM KH₂PO₄/K₂HPO₄, 8 mM (NH₄)₂SO₄, 2 mM citrate, 1 mM MgSO₄, 0.1 mg mL^–1 of thiamine-HCl, 1 mM glycerol, and 40 μg mL^–1 of a mixture of threonine, methionine, leucine, and histidine) supplemented with antibiotics and inducers was used for the agar plate assay. Chemical solutions (50 mM serine or 50 mM MeAsp) were applied to the centerline of the plate and incubated at 4°C for 12–16 h to generate a chemical gradient before cell inoculation. Overnight cultures of different evolved strains were washed twice with tethering buffer and applied to the plate at a distance of 1.5 cm from the line where the chemical was inoculated. Plates were incubated at 30°C for 24–48 h. Spreading bias was quantified by measuring the distances from the inoculation point of the expanding colony to its edges up or down the chemical gradient using ImageJ software and taking their ratio.

Microfluidics chemotaxis assay

Cells were harvested by centrifugation at 4000 rpm for 5 min and washed twice with the tethering buffer. Methyl aspartate dissolved in the tethering buffer (50 mM) was adjusted to pH 7.0. The responses of E. coli cells to a concentration gradient were measured using a microfluidic device illustrated in S7A Fig. E. coli cells were added into the well of the device (labeled as ‘Cells’) at the final OD₆₀₀ of 1.2–2 and equilibrated for 20 min in the observation channel. Methyl aspartate solution was added into the other well (‘Compound’) and allowed to diffuse into the observation channel through a porous agarose membrane, and thereby establish a concentration gradient. Fluorescence microscopy using a Nikon Ti-E microscope system with a 20X objective lens was used to detect the fluorescence intensity of cells in the observation area. The cellular response was characterized by the fluorescence intensity in the observation region (300 μm × 200 μm). Data were analyzed using ImageJ (Wayne Rasband, National Institutes of Health, USA).

FRET assay

FRET assay was performed as described previously [41,42]. Bacterial cells were transformed with pVS88 plasmid (CheY-YFP, CheZ-CFP) for FRET experiment and were grown in TB (1% tryptone, 0.5% NaCl, pH 7.0) supplemented with 100 µg mL^-1 ampicillin and 50 µM isopropyl-β-D-thiogalactoside (IPTG) at 34°C and 275 rpm. Cells were harvested at OD₆₀₀ of 0.6 and washed twice with the tethering buffer (10 mM KH₂PO₄/K₂HPO₄, 0.1 mM EDTA, 1 μM methionine, 10 mM sodium lactate, pH 7.0). Strains were immobilized in a flow chamber and stimulated with chemoeffectors added or removed by flow. FRET measurements were performed on an upright fluorescence microscope (Zeiss Axio Imager.Z1). The YFP/CFP ratio, calculated as the ratio between fluorescence intensities in YFP and CFP channels, was corrected for the baseline drift and normalized to the value at the beginning of an experiment. The pathway activity (kinase activity) was calculated from the YFP/CFP ratio assuming full activation (A = 1) by 300 µM NiCl₂ and full inactivation (A = 0) by the mix of 100 µM MeAsp and 50 µM serine, and fitted to a Hill function, A(L) = A₀ × (1 – L^H/(L^H + K_D^H)), where A(L) is the amplitude of the measured FRET response, A₀ is the pathway activity in the unstimulated case, L is the attractant concentration, H is the Hill coefficient and K_D is the half-inhibitory concentration. In case of R1 strain, an additional parameter accounting for non-zero activity upon high methyl aspartate stimulation was introduced.

Promoter activity assays

For promoter activity assays, E. coli strains, transformed with the reporter plasmid pAM109, were grown in TB supplemented with 50 µg mL^-1 kanamycin in 96-well plates at 30°C in a rotary shaker at 180 rpm. Cell fluorescence was measured using flow cytometry on BD LSR Fortessa SORP cell analyzer (BD Biosciences).

Motility analyses

For motility analysis, E. coli cells were grown in 10 mL TB medium at 34°C in a rotary shaker at 275 rpm. Cells (1 mL) at mid-log phase (OD₆₀₀ = 0.6) were collected and re-suspended in 1 mL motility buffer (6.15 mM K₂HPO₄, 3.85 mM KH₂PO₄, 100 μM EDTA, 67 mM NaCl, pH 7.0, 0.01% Tween80, and 1% glucose if not otherwise stated), and then diluted in motility buffer supplemented with indicated amounts of MeAsp to an OD₆₀₀ appropriate for cell tracking (~0.02-0.03). Swimming velocity and tumbling rate were measured by cell tracking in a glass chamber using phase-contrast microscopy (Nikon TI Eclipse, 10X objective NA = 0.3, CMOS camera EoSens 4CXP, 1 px = 0.7 µm). All data were analyzed using ImageJ with custom-written particle-tracking pipeline (https://gitlab.gwdg.de/remy.colin/particletracking2) for swimming velocity, drift velocity, and tumbling rate analysis [60]. In short, detection, via intensity thresholding, and tracking, via next-frame closest proximity, of single cells in a background-subtracted movie yields a set of 2D trajectories . Further analysis ignores trajectories shorter than 1 s and separates swimmer from non-swimmer trajectories via a criterion on diffusivity, µm²/s for swimmers. This threshold was chosen closely above the diffusion coefficient of non-swimmers, ~ 0.4 µm²/s at our low densities [61], to enable the analysis of tumbly mutants. Instantaneous swimming speeds of swimmers are measured via a local fit on a 10-frame sliding window. Swimmer trajectories are split between runs and tumbles based on a criterion on the local ballisticity of the trajectory, with calculated on the same sliding window as (normalized threshold ). We then use the fraction of time spent tumbling, i.e., the tumbling bias, , where sums the instances of state X within a trajectory (for computing distributions) or over all trajectories in a movie (for average values), to quantify tumbling. When building distributions of tumbling bias, each trajectory is weighted by its duration.

Chemotactic drift measurements and run analysis

Chemotactic drift and bias were measured according to the protocol previously described in detail [60]. Cells were grown in TB at 34°C in a rotary shaker at 275 rpm, harvested at OD₆₀₀ = 0.5 and washed three times in the motility buffer. Two chambers of a polydimethylsiloxane (PDMS) microfluidic device were filled with cells in the motility buffer (6.15 mM K₂HPO₄, 3.85 mM KH₂PO₄, 100 μM EDTA, 67 mM NaCl, pH 7.0, 1% glucose, 0.01% Tween80) supplemented either without or with 50 μM, 100 μM, 500 μM or 1 mM of MeAsp. Trajectories of single cells were recorded in the middle of the channel between the chambers using phase-contrast microscopy (Nikon TI Eclipse, 10X objective NA 0.3, CMOS camera EoSens 4CXP) at the acquisition rate of 50 frames per second and were further analyzed with the custom ImageJ particle-tracking pipeline (https://gitlab.gwdg.de/remy.colin/particletracking2). The average chemotactic drift velocity was calculated as , where and are the mean velocity in the direction of the gradient (y) and the trajectory length for each detected cell .

For the analysis of run durations, runs shorter than 5 frames (0.1 s) were removed and the direction of each run was calculated as , where and . All runs with |α| < π/4 and 3π/4 < |α| < π were classified as up and down the gradient, respectively (S8A Fig). The distributions of run durations for both subpopulations were visualized using MATLAB.

Simulations

Numerical simulations are described in details in S1 Text. The code is available at https://github.com/croelmiyn/ExperimentallyEvolvedStrains.

Statistical analysis

All experiments were done in triplicate. Error bars indicate respective standard errors of the mean. Significance analysis was done in comparison to the reference strains, as indicated in figure legends. P values were calculated using one-tailed and two-tailed Student t-test. The values for N, P, and the specific statistical test performed for each experiment are included in the appropriate figure legend.