Native regulation of motility genes in nutrient-rich medium maximizes swimming while limiting the cost of expression
To investigate how motility and growth depend on the expression of flagellar genes, we engineered a derivative of E. coli K-12 strain MG1655 with titratable expression of the flhDC operon that encodes the master activator of the entire flagellar regulon (Fig. 1a, Supplementary Table 1 and Methods). Expression of the flagellar regulon at different levels of Ptac-flhDC induction was quantified using a fluorescent reporter for flagellin (fliC gene) promoter activity (PfliC), which was previously shown to efficiently report the production of flagella in E. coli20,21,26. Reporter activity was measured using either a plate reader to follow changes in in the mean expression over time (Extended Data Fig. 1a, b), or flow cytometry to determine the distribution of single-cell expression levels within the cell population at a defined time point in mid-exponential phase (Fig. 1b). We confirmed that both readouts yielded similar results for E. coli cultures grown in nutrient-rich tryptone broth (TB) medium, with native MG1655 (wild-type; MG1655 WT) expression falling at an intermediate level within the range covered by the inducible Ptac strain (Fig. 1c and Extended Data Fig. 1b).
To understand how motility changes as a function of gene expression, we characterized swimming behavior in populations of MG1655 WT and Ptac cells using differential dynamic microscopy27 (see Supplementary Note 1 and Extended Data Fig. 2). We observed that population-averaged cell swimming velocity initially increased with expression at low levels of induction, but saturated at high levels of expression (Fig. 1d). Notably, this saturation occurred around the level of motility gene expression seen in the wild-type strain. A similar pattern was observed when the fraction of well-swimming cells within the population, as determined by our motility assay, and the swimming velocity of only these cells were plotted individually (Extended Data Fig. 1c, d). The cell swimming velocity at the highest expression level was even slightly reduced (Fig. 1d and Extended Data Fig. 1c). Two other derivatives of E. coli K-12, W311028 and RP437 (the latter is commonly studied as a wild type for E. coli chemotaxis29), both showed a similar relation between flagellar gene expression and motility, but were slightly less motile than MG1655 WT (Fig. 1d). The poorer swimming performance of RP437 may be a consequence of its extensive mutagenization29, and a previous study showed that the motility of this strain can be improved by experimental evolution20.
We further investigated the effect of motility on fitness by co-culturing CFP-labeled MG1655 WT or Ptac strains with a non-flagellated YFP-labeled ΔflhC strain. The fitness cost of flagellar regulon activity over a culture passage was determined as the reduction in relative cell number of the tested strain in the co-culture from the initial 50% at inoculation20,21. This cumulative fitness cost gradually increased with the level of motility genes expression over the entire range of induction tested (Fig. 1e). Thus, expression of motility genes beyond the native level in E. coli K-12 strains does not appear to provide any additional benefit, but nevertheless imposes an increasing fitness cost.
Hydrodynamic constraints limit cell velocity at high levels of flagellar production
The saturation of E. coli motility at high levels of flagellar gene expression could be due either to some bottleneck in the biogenesis of functional flagella or to limits in the physical propulsion by multiple flagella. To distinguish between these two possibilities, we first determined how the activity of the flagellar regulon corresponds to changes in flagellation. Staining flagella with an amino-specific fluorescent dye30 revealed a clear dependence of the number and length of flagella on the expression of the flagellar regulon (Fig. 2a). The average number of flagellar filaments per cell showed an approximately linear increase with the activity of the PfliC reporter (Fig. 2b, Extended Data Fig. 3a). The length of flagellar filaments also showed a moderate increase followed by an apparent saturation (Fig. 2c, Extended Data Fig. 3b). These results were consistent with increased amounts of intra- and extracellular flagellin, determined by immunoblotting (Extended Data Fig. 4). Thus, E. coli cells can synthesize more flagella at levels of motility gene expression that exceed those of wild-type cells, but this increase does not translate into higher swimming velocity.
Alternatively, this saturation of swimming with flagella number could be explained by the physics of E. coli motility. The hydrodynamics of flagella-propelled bacterial swimming is well understood and can be captured by relatively simple mathematical models such as resistive force theory (RFT)31,32. We therefore used RFT to describe the swimming of a multi-flagellated bacterium, where multiple flagella form a tight bundle that rotates to propel the cell (Supplementary Note 2 and Extended Data Fig. 5a). Based on our experimental measurements (Extended Data Fig. 5b,c), we assume that the flagellar motors operate at a constant speed that does not depend on the number of flagella, which may be the maximum speed of the motor torque-speed relationship. Indeed, the load per motor is low and decreases as the number of flagella increases (Extended Data Fig. 5g), because now multiple motors share the torque generation necessary for bundle rotation and cell propulsion. Our model predicts that swimming velocity should initially increase with motility gene expression and then saturate, in agreement with the experimental data (Fig. 2d and Extended Data Fig. 5d-f). The initial increase stems from the increase of flagellar length and the increased thickness of the bundle formed by more flagella. Saturation then occurs in the RFT model at high number of filaments because the viscous drag of the cell body becomes negligible compared to the drag of the flagella themselves. As a consequence, any increase in thrust resulting from adding more flagella is offset by an equal increase in viscous drag, since the two have identical dependencies on flagellar length and bundle thickness. Although our model is clearly simplified, in particular, does not capture all the complexity of flagella bundle hydrodynamics33, it strongly indicates that the ability of E. coli to increase its swimming velocity by increasing the number and length of flagella is indeed limited by the hydrodynamics and mechanics of flagellar propulsion in viscous media.
Motility gene expression follows the potential benefit of chemotaxis under carbon-limited conditions
Since expression of the flagellar regulon is under catabolite repression during carbon-limited growth in minimal media, we asked whether this regulation serves to maximize swimming, as observed in nutrient-rich medium, or whether it optimizes an alternative target. Consistent with its C-line-dependent regulation7,21,25, the expression of motility genes in the MG1655 WT strain grown in the minimal medium was much lower in the presence of a good (glucose) than a poor (succinate) carbon source (Fig. 3a). Expression in the Ptac strain at a given induction was also lower during growth on glucose, but this dependence was weaker, as expected for promoters that are not catabolite repressed34. Despite these differences, both swimming velocity (Fig. 3b) and growth fitness cost (Extended Data Fig. 6) in the Ptac strain showed the same dependence on motility gene expression for both carbon sources. MG1655 WT levels also fit to this curve, but unlike growth in nutrient-rich medium, the native activity of the flagellar regulon clearly does not maximize swimming velocity in this case.
Instead, we hypothesized that native gene expression under carbon-limited growth might correlate with the potential benefit that could be achieved in a given carbon source by performing chemotaxis towards sources of additional nutrients, as proposed before21. Following this previous study, we measured the benefit of chemotaxis by providing localized sources of amino acids in co-culture between the Ptac strain (labeled with CFP) and its motile but non-chemotactic ΔcheY derivative (labeled with YFP) for different levels of motility gene induction (Extended Data Fig. 7). While the benefit of chemotaxis saturated at high levels of motility gene expression in both carbon sources, saturation occurred at much lower expression in the presence of glucose, with the point of saturation close to the native level of expression in the respective carbon source.
Another notable finding was the appearance of two distinct subpopulations, with almost negative and strongly positive expression, at low average levels of reporter activity in the Ptac strain (Fig. 3a and Extended Data Fig. 8). Interestingly, this separation appeared to be a function of the average reporter activity and did not depend on the carbon source (Fig. 3a and Fig. 3c). In this low expression range, the proportion of positive cells in the population increased up to a critical level of expression, after which the distribution became unimodal and it was rather the mean of the positive peak that increased with induction. Motility gene expression in MG1655 WT cells was above the critical level where bimodal behavior becomes apparent, even in culture grown on glucose. To investigate whether native regulation could also exhibit bimodality, we further reduced motility gene expression in wild-type cells by prolonged growth under catabolite repression in glucose, either by using a higher dilution of the TB-grown overnight culture or by pre-growing the overnight culture in glucose (Fig. 3c, Extended Data Fig. 9a). Indeed, both conditions reduced PfliC activity in the MG1655 WT cell population and revealed a bimodal pattern similar to that observed in the Ptac strain. Bimodality was also observed for a non-induced Ptac strain grown in TB (Fig. 1c and Fig. 3c). Thus, bimodality appears to depend solely on the expression level and not on the details of transcriptional regulation of the flhDC operon or on the growth medium.
Motility gene expression in E. coli has previously been shown to be pulsatile26,35 and this may be the cause of the observed bimodality. In the closely related species Salmonella enterica, motility genes are also known to exhibit bistable expression36. Both bistability (in S. enterica) and pulsatility (in E. coli) of expression were attributed to negative regulation of FlhDC activity by YdiV (RflP)37, with organism-specific differences in the topology of the YdiV regulatory circuit35,37. We therefore tested whether regulation by YdiV could be responsible for the emergence of bimodality in our experiments. As expected, the expression level of motility genes in a ΔydiV strain was elevated, and it was above the bimodality threshold in glucose even when the culture was inoculated from TB at a 1:1000 dilution (Fig. 3c and Extended Data Fig. 9b). However, when the expression level was sufficiently lowered by pre-growth in glucose, two distinct subpopulations could be clearly observed in the ΔydiV strain, suggesting that negative regulation by YdiV is not sufficient to explain the bimodal activation of the PfliC reporter.
Activity of the flagellar regulon in natural isolates of E. coli
Finally, to investigate how investment in motility varies among E. coli strains that may have adapted to different ecological niches, we used the ECOR collection, which contains 72 isolates from different hosts and geographical regions38. From this collection, we first selected 61 strains that were sensitive to kanamycin and thus transformable with the PfliC reporter plasmid, and then discarded 23 non-swimming isolates that did not spread in porous (0.27%) TB agar. From the remaining 38 spreading isolates, a subset of 24 strains with moderate and good spreading abilities was chosen for further investigation (Supplementary Table 2).
Although the activity of the PfliC reporter varied widely among the TB-grown ECOR strains, it was consistently below or similar to that of the MG1655 WT strain (Fig. 4a and Extended Data Fig. 10a), indicating that the investment in motility by natural E. coli isolates is under similar limitation as in the K-12 strains. However, the swimming velocity of the majority of ECOR strains grown in liquid TB medium was lower than that of MG1655 WT and Ptac strains at similar levels of PfliC reporter activity (Fig. 4a and Extended Data Fig. 10a). Since previous studies showed that the motility of several pathogenic E. coli strains39 and other bacteria40 can be activated when cells are grown on a surface or in a porous medium, we measured the ability of ECOR strains to spread in porous 0.27% TB agar. Indeed, the spreading of most ECOR strains, including those that were poorly motile when grown in liquid, was comparable to that of MG1655 WT and Ptac (Fig. 4b).
A possible explanation for this difference could be increased expression of motility genes in cells grown in porous media or on a semi-solid agar surface, where flagella rotate under high load39,41–43. We therefore measured the activity of the PfliC reporter in cultures grown on 0.5% TB agar plates. In this case, expression in individual strains correlated well with their spreading (Extended Data Fig. 10b). While we indeed observed an upregulation of reporter activity in such surface-grown compared to liquid-grown cultures for a few isolates (e.g. ECOR-72), this was not the case for the majority of ECOR strains (Fig. 4c, Extended Data Fig. 10c and Supplementary Table 2). However, when the motility of cells grown on an agar surface was subsequently analyzed in motility buffer (see Methods for details), the average cell swimming velocity was indeed higher for many ECOR strains compared to liquid-grown cultures, now showing a dependence of swimming velocity on expression similar to the MG1655 WT and Ptac strains (Fig. 4d, Extended Data Fig. 10d and Supplementary Table 2). Thus, the observed poor motility of many ECOR isolates grown in liquid medium cannot be generally explained by low activity of the flagellar regulon but rather indicates some deficiency in flagellar assembly or function in liquid-grown cell. Notably, however, both motility gene expression and swimming of all ECOR strains were always below or comparable to that of MG1655 WT, further supporting the fundamental nature of limitation imposed on E. coli motility by hydrodynamics.