The E. coli K-12 wild-type strain NCM3722 is a non-motile strain (Cremer et al., 2019) (Supplementary Figure S1) since it carries a lone missense mutation (N87K) in the gene of fliC which encodes flagellin (Lyons et al., 2011), possibly affecting the function of its flagella. Thus, this mutated fliC of NCM3722 was complemented to the wild-type fliC of another widely used E. coli K-12 wild-type strain MG1655, resulting in the parent strain, CY598. With the complemented FliC, CY598 was weakly motile on the soft agar plate (Supplementary Figure S1). In order to obtain hypermotile mutants, CY598 was inoculated into a soft agar plate, and the appearance of swimming zones was recorded at different time points (Figure 1A). After incubation for 120 h, big swimming zones appeared (Figure 1A). Two strains (CY706 and CY708) were isolated from different areas of the swimming zones, and their motility was tested. When compared to the parent CY598, they showed 2–3 folds higher swimming motility in soft agar plates (Figure 1B) and approximately 10-fold faster swimming speed under the microscope (Figure 1C, Supplementary Videos S1–S3). The flagella produced by these strains were further observed under the transmission electron microscope (TEM) (Figure 1D). CY706 and CY708 produce a much larger number of flagella than the parent CY598, supporting the hypermotility of the two strains. Then, we investigated the expression level of genes involved in motility by quantitative real-time PCR (qRT-PCR). The expression of several example genes in Class I, II, and III of the entire flagellar cascade was tested (Figure 1E). The genes in Class II and III showed >30 folds increased expression in CY706 and CY708 as compared to that in CY598. However, there were no significant expression changes of flhDC, encoding the master regulator of motility regulon in Class I. Consequently, we assessed the protein amount of FlhD by western blots and found that the FlhD protein was highly accumulated in both CY706 and CY708 (Figure 1F). Thus, the accumulation of FlhD protein in these evolved strains might mediate the hypermotile phenotype observed.

To find out the possible reasons that induced the accumulation of FlhD in CY706 and CY708, we performed whole-genome sequencing of CY598, CY706, and CY708. As compared to the genome of CY598, we found that CY706 and CY708 shared one common point mutation. This common G232T mutation located at the 232nd nucleotide in the clpX gene, which leads to the change of amino acid in the ClpX protein from valine to phenylalanine at position 78, giving the mutated ClpX^V78F (Figure 2A). This V78F mutation was located in the large AAA+ domain of ClpX (Stinson et al., 2013) (Figure 2B). The shared ClpX^V78F mutation might explain the highly motile phenotype mediated by the accumulation of FlhD that we observed both in CY706 and CY708.

To verify this point, using the λ Red system (Datsenko and Wanner, 2000), we replaced the wild-type clpX gene in the parent strain CY598 by the mutated clpX (G232T), resulting in the strain CY1102, and complemented the mutated clpX (G232T) in the hypermotile strain of CY708 with the wild-type clpX gene in parallel, resulting in the strain CY1166. Then, their motility, flagella, and the amount of FlhD protein were individually assayed. Similar to CY708 and CY706 (Figures 1B,C), CY1102 carrying ClpX^V78F was highly motile (Figure 3A, Supplementary Video S4) and produced larger number of flagella (Figure 3B). In contrast, CY1166 carrying wild-type ClpX was weakly motile (Figure 3A, Supplementary Video S5) and produced less flagella (Figure 3B), exhibiting a similar phenotype as the parent CY598 (Figures 1B,C). In consistence with these observations, FlhD was accumulated highly in CY1102 to a similar level as that observed in CY708, but no accumulation of FlhD was observed in CY1166 (Figure 3C). At the same time, FlhC of the FlhDC regulator was also identified to be highly accumulated in the ClpX^V78F background (Supplementary Figure S2). These evidences suggested that it was the mutated ClpX^V78F that mediated the accumulation of FlhDC and enhanced the motility. Moreover, we monitored the protein level of FlhD in the ClpX null background because the FlhDC complex was highly accumulated in the absence of the ClpXP protease (Tomoyasu et al., 2003; Kitagawa et al., 2011). Interestingly, the amount of FlhD was increased to a similar level in the ClpX null background as that observed in the ClpX^V78F background (Figure 3C). Thus, the V78F mutation in ClpX seems to inactivate the ability of the ClpXP to hydrolyze FlhD completely. All lines of evidences suggested that the clpX ^G232T mutation contributed to the hypermotile phenotype observed.

People had reported that single-point mutations in clpX could enhance the motility in E. coli (Fraebel et al., 2017; Liu et al., 2019). Although the reported influence of these mutations on bacterial motility was much less than the enhancement induced by the ClpX^V78F mutation identified here, these observations indicated that the gene of clpX could be a hot locus to boost bacterial motility.

To further explore this point, applying the same method (Figure 1A), we performed a second round of selection for the hypermotile mutants. Intriguingly, more isolates carrying different point mutations in the clpX gene were characterized. Out of the total 18 hypermotile strains we isolated, four mutants carried the same Q289* (C865T) mutation (* indicates stop codon) and one mutant carried an L317R (T950G) mutation (Supplementary Figure S3). Thus, these evidences suggested that the gene of clpX could be a hot locus mutated spontaneously to boost bacterial motility.

How would these mutants induce the hypermotile phenotype? In the case of the ClpX carrying the Q289* mutation (Supplementary Figure S3), a truncated ClpX without the hinge and the small AAA+ domain would be generated (Stinson et al., 2013), which would lose its ability to degraded FlhDC. In the case of the ClpX carrying L317R mutation, the residue of L317R was localized in the hinge domain of ClpX (Supplementary Figure S3) (Glynn et al., 2012). Since leucine and arginine show distinct biochemical features with one hydrophilic and the other hydrophobic, the L317R mutation would affect the protease function of ClpX by changing the hinge structure. However, in the case of the ClpX carrying V78F mutation, it was not straightforward to predict the effect that this mutation would bring to the function of the mutated ClpX. The residue of V78F was located in the large AAA+ domain of ClpX (Stinson et al., 2013) (Figure 2). Since both valine and phenylalanine are hydrophobic amino acids except that they carry different sizes of side chains, it is quite a mystery how this V78F mutation would inactivate the ability of the ClpXP to hydrolyze FlhD (Figure 3C). Thus, the following study was to discover the molecular mechanism of ClpX^V78F mediated high accumulation of FlhDC.

We had shown that the FlhD in the ClpX^V78F background was accumulated to a similar level as that in the ClpX null background (Figure 3C). What could be the reasons? First, one possibility is that the protein of ClpX^V78F is unstable so that it makes strains carrying ClpX^V78F as a ClpX null strain. This hypothesis was ruled out by the observation that the protein level of the two types of ClpX proteins showed no significant difference between the ClpX^V78F mutants and the strains carrying the wild-type ClpX (Supplementary Figure S4). Next, ClpXP is an ATP-dependent protease (Fei et al., 2020). The structure study indicated that ATP bound to ClpX via contacting the side chains of V78 and I79 (Stinson et al., 2013). When both V78 and I79 were mutated into A with a very smaller side chain, the mutated ClpX required a higher concentration of ATP to support ATP hydrolysis and protein degradation in vitro (Stinson et al., 2013). Will the single V78F point mutation carrying a bigger side chain affect the ATP binding to ClpX by itself? To answer this question, using the I-TASSER server (Yang and Zhang, 2015), we modeled the structure of ClpX^V78F by specifying chain A of the reported wild-type ClpX structure (PDB code 4I81) as the template. Compared to V78 in the wild-type ClpX (Figure 4A), the F78 in the mutated ClpX was much closer to the bound ATPγS (Figure 4B). The bulky aromatic side chain of F78 protruded into the ATP-binding pocket and occupied a substantial part of the pocket (Figures 4C,D). As a result, it will be difficult for ATP to bind to the ClpX^V78F due to the steric hindrance effect produced by the bulky side chain of the F78 residue. Thus, the mutated ClpXP might lose its ability to degrade FlhDC.

We further purified ClpX^WT and ClpX^V78F and compared their in vitro ability of protein degradation using FlhD as the target. We found that the wild-type ClpXP protease degraded FlhD effectively at the presence of either 2.5 millimolars (mM) or 5 mM ATP (Figure 4E, Figures 4G–H). As expected, it showed higher degradation efficiency with a higher concentration of ATP. In contrast, the degradation activity of the mutated ClpXP protease was largely impaired (Figure 4F, Figures 4G–H). No FlhD degradation was observed at 2.5 mM ATP, but a partial activity of FlhD degradation existed at 5 mM ATP, indicating a higher ATP concentration could slightly restore the protease function of the mutated ClpXP. These results indicate that the V78F mutation affected the affinity of ClpX^V78F to ATP and, therefore, the activity of the mutated ClpXP protease during degradation of FlhD, which is consistent with the prediction described above using the structure-modeling methods. In other words, ClpX^V78F is more sensitive to deprivation of ATP than ClpX^WT, at least during degradation for certain targets like FlhD. Given that under most environmental conditions, the ATP level within bacterial cells is limited and tightly tuned (Chapman et al., 1971; Sielaff et al., 2018; Meyrat and von Ballmoos, 2019), and a single-point mutation such as ClpX^V78F could extensively affect the stability of many proteins and physiological processes.

Moreover, this ClpXP mutant might fail to hydrolyze other protein targets accordingly. To verify this point, we integrated a global proteome study with a transcriptome study to investigate how the other targets of ClpXP changed among the ClpX^V78F mutants, and the ClpX null and wild-type strains. A target of ClpXP protease was expected to show changes at the protein level but not at the transcription level. Including FlhD and FlhC, eight targets of ClpXP protease (Flynn et al., 2003; Nagashima et al., 2006) were identified to show significant changes solely at the protein level (Figure 5A). As expected, the FlhD protein accumulated in the strains carrying ClpX^V78F to a level comparable to that in the ClpX null strain, which fitted to the previous data of the western blots well (Figure 3C). Interestingly, the other five proteins, including FlhC, RseA, RecN, MoaA, and ZapC, were changed similarly as the FlhD protein among these strains. But two of them (RpoS and YbaQ) were affected differently as their protein amount accumulated more in the ClpX null background when compared to the ClpX^V78F background. Given that the bulk of the targets of ClpXP protease identified in this work showed similar changes in the ClpX null background and in the ClpX^V78F background, this supported our hypothesis that this point mutation of ClpX^V78F inactivated most of its protease activity via affecting ATP binding. However, the mutated ClpXP protease could still retain slight residue ability to degrade some targets possibly including RpoS and YbaQ as we identified here, indicating ClpXP protease could have varied specificity to its targets.

Although FlhDC is reported to be degraded by ClpXP, the degradation signal recognized by ClpXP has never been reported. Since FlhC could not be degraded by ClpXP without forming a complex with FlhD (Takaya et al., 2012), FlhD might contain the degradation signal. Since the C-terminal five amino acid residues (RKKRA) of FlhD are highly similar to the reported degradation signals (RKKAI) of the other proteins recognized by ClpXP (Flynn et al., 2003), the C-terminal RKKRA of FlhD could function as the signal tag for ClpXP recognition. Indeed, when we deleted the C-terminal five amino acid residues, the truncated FlhD accumulated in the ClpX wild-type background to a similar level as that observed in the ClpX^V78F and ClpX null mutants (Figure 5B). Additionally, the amount of wild-type FlhD accumulated in the ClpX^V78F background was comparable to the amount of the truncated FlhD detected in the ClpX^WT or ClpX^V78F background (Supplementary Figure S5). Therefore, the C-terminal RKKRA of FlhD functioned as the signal tag for ClpXP recognition. As a result, ClpXP directly regulates FlhDC via hydrolysis in E. coli K-12 strains in vivo in a way similar to that it degrades other known protein targets.

Finally, we analyzed the transcriptome of these hypermotile (CY706, CY708, CY1102) and weakly motile strains (CY598, CY1166) systematically. In this process, we could reveal the FlhD or FliA regulons by identifying genes that were expressed similarly as the flagella genes. In parallel, since flagella synthesis is energetically expensive (Soutourina and Bertin, 2003; Fontaine et al., 2008), flagella synthesis in the hypermotile mutant might affect the biosynthesis of other processes and the corresponding genes, and our global transcriptome study could discover those genes. Indeed, we obtained both up- and down-expressed genes in the hypermotile strains. In total, 114 genes were identified as differentially expressed, with 85 genes being upregulated and 29 genes being downregulated in the hypermotile strains (Supplementary Table S1, Figure 6A).

As expected, the bulk of these upregulated genes were reported to be regulated by FlhDC in a direct or indirect mode, including the 51 flagella genes and genes encoding phosphodiesterases which hydrolyze c-di-GMP (Figure 6B). We further identified that four additional genes (argI, lgoT, rcsA, and yahA) were also regulated by FlhDC in a direct or an indirect mode since their expression was reduced significantly when FlhDC was deleted (Supplementary Figure S6). We noted that the gene of yahA (pdeL) encodes a c-di-GMP-specific phosphodiesterase. The FlhDC-regulated PdeH c-di-GMP phosphodiesterase was well-studied which maintains the low c-di-GMP level in order to promote motility (Guttenplan and Kearns, 2013; Serra and Hengge, 2019). The PdeL identified here could be another phosphodiesterase controlled by FlhDC directly or indirectly which works together with PdeH to control the level of c-di-GMP (Reinders et al., 2016).

Interestingly, a bulk of these downregulated genes functioned in biofilm formation, including bhsA, qseBC, and ecpB (Zhang et al., 2007; Garnett et al., 2015; Li et al., 2020), and functioned in stress response resistance, especially acid stress, including gadW, gadX, hdeD, slp, ygiW, uspD, and iraP (Nachin et al., 2005; Burton et al., 2010; Krin et al., 2010; Lee et al., 2010; Seo et al., 2015; Dorich et al., 2019) (Supplementary Table S1, Figure 6B). As a result, genes involved in the bacterial motility and genes related to biofilm formation and stress response seemed to show antagonistic expression.

To test if this antagonistic expression is a specific observation or a common phenomenon, we compared the transcriptome of two MG1655 variants (CY570 and CY562), with one hypermotile and the other one non-motile. The hypermotile MG1655 (CY570) carried an IS insertion upstream of the flhDC promoter (Supplementary Table S2 of strain list), and flhDC was highly expressed in the hypermotile MG1655 strain (Supplementary Figure S7). Approximately 80% of the 114 genes identified in the NCM3722 background were also differentially expressed between the hypermotile and the non-motile MG1655 strains (Supplementary Figure S8). Similarly, genes functioned in motility were upregulated, but genes related to stress resistance and biofilm formation were downregulated in the hypermotile MG1655 strain (Figure 6B, Supplementary Table S1). Thus, the antagonistic expression between motility genes and genes related to biofilm formation and stress response could be a common phenomenon in the K-12 strains of E. coli. Since the stress-related genes showed a bigger fold of change in the MG1655 background (Figure 6C), we compared the acid stress response of the non-motile and hypermotile MG1655 strains. Indeed, the hypermotile MG1655 was less resistant to the acid stress than the non-motile strain (Figure 6D).

As a result, in the hypermotile strains, cells reduced the expression of genes related to biofilm formation and stress response in order to reallocate more resources for the biosynthesis of flagella. Given the important role of stress response in bacteria survival (Schellhorn, 2020), these evidences showed molecular reasons for the reduced fitness of the hypermotile bacteria.

To construct plasmid encoding the N-terminal His-tagged ClpX, the open reading frame (ORF) sequence of E. coli ClpX^WT or ClpX^V78F was PCR-amplified as a fragment flanked by the NdeI and HindIII sequences using primers of His-clpX-F (5′-CAT ATG ACA GAT AAA CGC AAA GAT GG-3′) and His-clpX-R (5′-AAG CTT GGT TAA TTA TTC ACC AGA TGC CTG-3′), and sub-cloned into plasmid pET28a. The resulted plasmid was designated as pCY519 (for ClpX^WT) or pCY524 (for ClpX^V78F).

To construct plasmid encoding the C-terminal His-tagged ClpP, the ORF sequence of E. coli ClpP was PCR-amplified as a fragment flanked by the NheI and XhoI sequences using primers of His-clpP-F (5′-GCT AGC TCA TAC AGC GGC GAA CGA GAT-3′) and His-clpP-R (5′-CTC GAG ATT ACG ATG GGT CAG AAT CGA-3′), and sub-cloned in plasmid pET24a. The resulted plasmid was designated as pCY526.

To construct plasmid encoding the N-terminal His-tagged FlhD, the ORF sequence of E. coli FlhD was PCR-amplified as a fragment flanked by the NdeI and HindIII sequences using primers of His-flhD-F (5′-CAT ATG CAT ACC TCC GAG TTG CTG AA-3′) and His-flhD-R (5′-AAG CTT TCA TGA TCA GGC CCT TTT CTT G-3′), and subcloned in plasmid pET28a. The resulted plasmid was designated as pCY531.

The E. coli strains used and constructed in this study are listed in Supplementary Table S2. This mutated fliC of NCM3722 (Lyons et al., 2011) was complemented to the wild-type fliC of another widely used E. coli K-12 wild-type strain MG1655 via P1 transduction. In the genome, fliC locates between fliA and fliDST operon. To avoid the possible polar effect generated by adding a selection maker to fliC, we added a kanamycin marker to the downstream of fliT. In brief, the kan fragment encoding a kanamycin marker from the plasmid pKD4 was PCR-amplified and integrated into the chromosome right downstream of the fliT in MG1655 via the λ Red system (Datsenko and Wanner, 2000). The fliT-kan marker and fliC in MG1655 were transferred into NCM3722 via P1 transduction mediated by phage P1vir, resulting in CY598. The fliC in CY598 was confirmed to be the functioning wild-type fliC by DNA sequencing.

To introduce the point mutation of clpX ^G232T (ClpX^V78F) into E. coli strains, the clpX ^V78F-cat allele was first amplified by overlap PCR, using genomic DNA isolated from the hypermotile mutant CY706 and the plasmid pBSK-lcml (Li et al., 2019) as the template for clpX ^V78F and cat, respectively. Subsequently, the clpX ^V78F-cat allele was integrated into the chromosome of target strains through the λ Red system (Datsenko and Wanner, 2000). Similarly, to complement the clpX ^V78F mutants, the clpX ^WT-cat allele was introduced in target strains in the same manner as described above, except that genomic DNA isolated from CY598 was used as the template for clpX ^WT. When necessary, the cat allele was removed by the Cre-LoxP system (Srividya et al., 2020).

To construct the ∆clpX mutant, the cat allele amplified from pBSK-lcml was introduced into target strains to replace the entire clpX gene in the chromosome, via the λ Red system (Datsenko and Wanner, 2000). When necessary, the cat allele was removed by the Cre-LoxP system (Srividya et al., 2020).

To construct a chromosomal fusion of C-terminal clpX-flag tag in E. coli strains, the clpX-flag-stop codon-cat allele was first obtained via overlap PCR amplification, using the plasmid pBSK-lcml as the template. Subsequently, the clpX-flag-stop codon-cat allele was integrated into the chromosome of target strains through the λ Red system (Datsenko and Wanner, 2000). The construction of C-terminal flhC-flag and flhD-tag-free-cat in E. coli strains was performed in the same manner as that described for clpX-flag.

To construct the ∆clpP mutant, the cat allele amplified from pBSK-lcml was introduced into target strains to replace the clpP gene in the chromosome, via the λ Red system (Datsenko and Wanner, 2000).

To construct the ∆flhDC mutant, the cat allele amplified from pBSK-lcml was introduced into target strains to replace the flhDC operon in the chromosome, via the λ Red system (Datsenko and Wanner, 2000).

All E. coli strains were grown in the N^-C^- minimal medium (You et al., 2013; Li et al., 2019) supplemented with 22 mM glucose (Sigma-Aldrich Corp., St. Louis, MO, United States), 20 mM NH₄Cl (Sigma-Aldrich Corp.), and 0.2% (w/v) casamino acid (Aladdin, Shanghai, China), except where indicated. When needed, 100 μg/ml ampicillin (Aladdin), 50 μg/ml kanamycin (Sangon Biotech, Shanghai, China), or 25 μg/ml chloramphenicol (Solarbio, Beijing, China) was supplied in the medium. Except where indicated, all growths were carried out in a 37°C water bath shaker with aeration as described previously (Li et al., 2019).

CY598 was grown to early exponential phase (OD₆₀₀ = 0.3–0.4) and spotted on soft agar plates [0.5% beef extract, 1% peptone, 0.5% NaCl (Solarbio), and 0.4% agar (Sigma-Aldrich Corp.)]. The seeded soft agar plates were incubated at 37°C till swimming zones appeared. Subsequently, bacteria from several different areas of the swimming zones were streaked on regular LB plates [LB broth (Solarbio) and 1.5% agar (Sigma-Aldrich Corp.)] to isolate putative hypermotile mutant(s).

Early exponential phase cultures of strains were aliquoted and incubated at 37°C on soft agar plates [0.5% beef extract, 1% peptone, 0.5% NaCl (Solarbio), and 0.25% agar (Sigma-Aldrich Corp.)], and the diameters of swimming halos were measured.

Bacterial strains were grown to early exponential phase (OD₆₀₀ = 0.3–0.4) and diluted to an OD₆₀₀ of approximately 0.002 with fresh culture medium in a six-well microtiter plate (polystyrene, tissue culture treated, flat bottom; Jet Bio-Filtration Co., Ltd., Guangzhou, China). A cover glass (18 square millimeters and 0.13–0.17 mm thick) was gently put on the top of the bottom surface of each well, providing a 15–20 micrometer (μm) deep channel. Movement of bacterial cells was observed using the 10 × phase-contrast objective of an inverted microscope (Nikon ECLIPSE Ti2) and videoed at 10.06 frames per second with a camera Nikon D5-Qi2. The trajectory of cells was analyzed using the software NIS-Elements AR (5.02.00). First, the video was cropped to limit the number of the analyzed cells, usually ranging from 30 to 80. Cell trajectories were captured via the program Cell Motility with the following settings: detect the targets as dark and round spots with various sizes; use 2.8–3.6 μm as the typical diameter for the targets; adjust the contrast ratio to subtract the background (eliminating debris or particles with a distinct density from the bacterial cells); check the box of “Add tracks”; check the box of “Delete trajectories with duration shorter than frames” and type 302 in the blank bar. Check the intact trajectory data listed after every run of the Cell Motility program, and if necessary, data for false (non-cell) targets could be further deleted manually. Finally, the darkness of the background could be adjusted by moving the LUTs bars, and trajectories were colored in a time-lapse manner for exported figures. To calculate the motility speed of a single cell, the same method was used as described above except for changing one of the settings in Cell Motility: check the box of “Delete trajectories with duration shorter than frames” and type 101 in the blank bar. Data of trajectories including path length (μm) and duration [second (s)] were exported, and the motility speed of a single cell was calculated as the path length divided by duration.

Strains were grown to early exponential phase (OD₆₀₀ = 0.3–0.4) and harvested by centrifugation at 4,500 rpm. The pellets were washed with sterile double distilled water (DDW) and finally resuspended in DDW to an OD₆₀₀ of approximately 1.0. 10 microliters (μl) of the resuspension described above was dropped on carbon-film-coated copper grids (400 mesh; Jingying Chemical Technology Co., Ltd., Guangzhou, China). After 15-min incubation at room temperature, fluid on the copper grids was absorbed with filter papers gently. Subsequently, 10 μl 0.6% (w/v) uranyl acetate was loaded on the copper grids to stain bacteria attached to the carbon film. After 1-min incubation at room temperature, fluid on the copper grids was absorbed with filter papers gently. Finally, copper grids with stained bacteria were completely dried at room temperature for at least 6 h. The flagella structure of bacterial strains on the copper grids was observed under transmission electron microscope HITACH HT7700.

The total RNA of strains was extracted as previously described (Li et al., 2019). First, 1 μg of total RNA was reversed to cDNA by using PrimeScript™ RT reagent Kit with gDNA Eraser (Takara Bio Inc., Shiga, Japan), according to the manufacturer’s instructions. Subsequently, qRT-PCR was performed with TB Green Premix Ex Taq™ II (Tli RNaseH Plus; Takara Bio Inc.). In each reaction, the final concentration of primers was 0.5 micromolars (μM), and 100-time diluted cDNA was used as the template. Primers used for qRT-PCR are listed in Supplementary Table S3.

30 ml of bacterial cells from the early exponential phase (OD₆₀₀ = 0.3–0.4) were harvested by centrifugation at 4°C and subsequently resuspended in 300 μl PBS. The resuspension was subjected to sonication, followed by 20-min centrifugation at 4°C for collecting the supernatant. The protein concentration was determined using the Enhanced BCA Protein Assay Kit (Beyotime, Shanghai, China). For each strain, the same amount of protein (∼50 μg) was loaded for SDS-PAGE electrophoresis. To detect target proteins, transferred membranes were incubated with corresponding antibodies and followed by treatment with the SuperSignal™ West Pico PLUS Chemiluminescent Substrate (Thermo Fisher, Waltham, MA, United States). The membranes were exposed to Tanon-6100C (Tanon, Shanghai, China), and western blot bands were quantified using the software ImageJ. For each assay, RpoB was considered as a loading control. The primary antibodies used in this study are anti-FlhD (produced by Genscript, Nanjing, China), anti-RpoB (BioLegend, San Diego, CA, United States), and anti-Flag (Abcam, Cambridge, United Kingdom).

Genome sequencing of E. coli strains in this study was performed using the combination of Illumina and Pacbio techniques. Total DNA was isolated from the overnight LB culture of each strain. In brief, for Illumina sequencing, the qualified DNA samples were processed to construct a library according to the following steps. 1) The genomic DNA was treated into fragments. 2) The staggered ends of the fragments were repaired to blunt ends, using T4 DNA polymerase, Klenow DNA polymerase, and T4 PNK. 3) Specific adaptors were added to the 3′-ends of the fragments. Finally, the qualified libraries were subjected to cluster preparation and sequencing. For Pacbio sequencing, the qualified genomic DNA was broken into fragments >10 kb with g-TUBE. Damages and ends of the fragments were then repaired, followed by flanking both ends with a hairpin adapter, resulting in a structure known as SMRTbell. Primers were annealed to SMRTbell, which attached the fragments to the bottom of the ZWM polymerase, the enzyme performing the final sequencing. Clean data from the sequencing processes above were mapped to the NCM3722 genome (NCBI accession number for chromosome: CP011495.1, for plasmid F: CP011496.1) for reference assembly.

To purify the wild-type and mutated ClpX proteins, 60 ml bacterial cells of strain CY1222 or CY1223 harboring pCY519 or pCY524 were grown in LB broth (Solarbio) supplemented with 5 mM MgCl₂ (Aladdin) and kanamycin at 37°C to an OD₆₀₀ of ∼0.8. 0.2 mM IPTG (Solarbio) was subsequently added to the cell culture, and the cells were harvested after incubation at 12°C for 24 h. The pellets were washed twice using buffer A [50 mM Tris-HCl, pH 7.5 (Solarbio), 100 mM NaCl (Solarbio)] and resuspended in 1.5 ml buffer B [buffer A, 5 mM imidazole (Solarbio), 5 mM MgCl₂]. The resuspended cells were subjected to a 40-min sonication program (every 5-s sonication was separated with a 5-s interval), and the supernatant was collected via 20-min centrifugation at 4°C. The supernatant was subsequently loaded to a gravity column containing Ni NTA Beads 6FF (Coolaber, Beijing, China), which had been balanced with buffer B in advance. The column was washed with buffer C (buffer A, 30 mM imidazole), and the proteins within the column were finally eluted using buffer D (buffer A, 250 mM imidazole). The elution containing ClpX was dialyzed against buffer E [buffer A, 10% glycerol (Macklin, Shanghai, China)].

To purify ClpP, 60 ml bacterial cells of strain CY1234 harboring pCY526 were grown in LB broth supplemented with kanamycin at 37°C to an OD₆₀₀ of ∼0.8. 0.2 mM IPTG was subsequently added to the cell culture, and the cells were harvested via centrifugation after incubation at 30°C for 3 h. The pellets were washed twice using buffer A and resuspended in 1.5 ml buffer F (buffer A, 5 mM imidazole). The resuspended cells were subjected to a 40-min sonication program (every 5-s sonication was separated with a 5-s interval), and the supernatant was collected via 20-min centrifugation at 4°C. The supernatant was subsequently loaded to a gravity column containing TALON Metal Affinity Resin (Takara), which had been balanced with buffer F in advance. The column was washed with buffer C, and the proteins within the column were finally eluted using buffer D. The elution containing ClpP was dialyzed against buffer E.

To purify FlhD, 120 ml bacterial cells of strain CY1279 harboring pCY531 were grown in LB broth supplemented with kanamycin at 37°C to an OD₆₀₀ of ∼0.8. 0.2 mM IPTG was subsequently added to the cell culture, and the cells were harvested via centrifugation after incubation at 30°C for 3 h. The same methods as described above for ClpP purification were used to obtain the FlhD protein, except that buffer G (buffer A, 60 mM imidazole) was used instead of buffer C during the wash step.

The assay was conducted as previously described but with slight modifications (Takaya et al., 2012). In brief, a proper amount of ClpX, ClpP, and FlhD was incubated together at 37°C in the reaction buffer, comprising 50 mM HEPES-KOH, pH 7.6, 50 mM KCl, 10 mM Mg (CH₃COO)₂, and 5 mM DTT, at the presence of 0, 2.5, or 5 mM ATP. 10 μl of the mix for each reaction was aliquoted at time 0, 1, and 2 h and finally subjected to SDS-PAGE electrophoresis and Coomassie-brilliant-blue staining, for detecting the amount change of proteins. Protein bands were quantified using the software ImageJ.

Bacterial cells from the early exponential phase (OD₆₀₀ of ∼0.4) were used for RNAseq assays. MG1655 strains were grown in the N^-C^- minimal medium (You et al., 2013; Li et al., 2019) supplemented with 22 mM glucose (Sigma-Aldrich Corp.) and 20 mM NH₄Cl (Sigma-Aldrich Corp.). RNAseq assays were performed as previously described (Li et al., 2019). Three independent total RNA extraction and RNAseq assays were performed for strains derived from NCM3722 and two independent total RNA extraction and RNAseq assays were performed for strains derived from MG1655. RPKM method (Reads Per Kilo bases per Million Reads) (Mortazavi et al., 2008) was used to calculate the expression of each gene. Gene expression in the hypermotile mutants (CY706, CY708, CY1102) was individually compared to that in the weakly motile strains (CY598, CY1166). In total, four comparisons were performed, including gene expression in CY706 (resp. CY708 or CY1102) versus gene expression in CY598 and gene expression in CY708 versus gene expression in CY1166. Genes that showed a p-value ≤ 0.05 in each comparison by t-test and with the average fold change of the four comparisons higher than 1.5 were defined as differentially expressed genes (DEGs). Since only one comparison of the hypermotile and non-motile MG1655 strains was obtained, DEGs between the two transcriptomes of the MG1655 strains were characterized by edgeR (Robinson et al., 2010; Rajkumar et al., 2015) with 1.5-fold cutoff and adjusted p-value (q-value) < 0.01.

300 ml of bacterial cells from the early exponential phase (OD₆₀₀ = 0.3–0.4) were mixed with precool (from −80°C) 60% methanol (v/v) and harvested by centrifugation at 4°C. Cell pellets were washed with PBS, and the cells were collected again by centrifugation at 4°C. To perform proteomic analysis, the cells were first resuspended in four volumes of lysis buffer [8 molars urea (Sigma-Aldrich Corp.) and 1% protease inhibitor (Calbiochem, Merck, Darmstadt, Germany)]. The resuspension was subjected to sonication on ice using an ultrasonic processer (Scientz, Ningbo, China), followed by 10-min centrifugation at 4°C, to collect the supernatant and discard debris. The total protein concentration was determined by using the PCA kit. The total protein was subjected to 30-min reduction using 5 mM dithiothreitol (Sigma-Aldrich Corp.) at 56°C and subsequently incubated with 11 mM iodoacetamide (Sigma-Aldrich Corp.) at room temperature in darkness for 15 min. The protein solution was then diluted with 100 mM TEAB (Sigma-Aldrich Corp.), to alter the urea concentration to less than 2 molars. After the treatments above, the protein was digested overnight with trypsin (Promega, Madison, WI, United States) at a mass ratio of 1:50 (trypsin to protein) at 37°C, followed by 4-h digestion at a mass ratio of 1:100 (trypsin to protein).

The peptides from digestions above were dissolved in Solvent A [0.1% formic acid (Fluka, Sigma-Aldrich Corp.)] and separated by using the NanoElute UPLC system. In the process, the gradient of Solvent B [0.1% formic acid in 98% acetonitrile (Fisher Chemical, Thermo Fisher)] was set at the following series: Time 0–44 min, 6–22%; Time 44–54 min, 22–35%; Time 54–57 min, 35–80%; Time 57–60 min, 80%. This was conducted on the UPLC system at a flow-rate of 300 nanoliter/min. The separated peptides were ionized in a capillary ion source and subsequently analyzed by using timsTOF Pro mass spectrometry. The voltage used for the ion source was 1.6 kilovolt. The parent ion and its secondary fragments of the peptide were detected and analyzed by TOF. The scanning range of the secondary MS was set at 100–1,700 M/Z, and the data acquisition mode of Parallel Accumulation-Serial Fragmentation (PASEF) was used. After the collection of the primary mass spectrum, the secondary spectrum with the charge number of the parent ion in the range of 0–5 was collected in PASEF mode for 10 times. The dynamic exclusion time of tandem mass spectrometry scanning was set to 30 s to avoid the repeated scanning of the parent ion. Data from the mass spectra described above were searched using MaxQuant (v1.6.6.0) throughout the provided transcriptome database (Genome of NCM3722, NCBI accession number for chromosome: CP011495.1, for plasmid F: CP011496.1). In the meanwhile, the reverse decoy database was concatenated with the transcriptome database, to calculate FDR. In addition, databases of common contaminator peptides were used for analysis, to reduce the effect of contaminations. Several other key settings are listed here: Trypsin/P was set as the cleavage enzyme, and the number of missing cleavages was not allowed to be bigger than 2; the mass tolerance set for precursor ions in First search and Main search was both 40 ppm, and the mass tolerance set for fragment ions was 0.04 Da; the fixed modification was set as carbamidomethyl on Cys, and the variable modifications were set as oxidation on and acetylation on N-terminus of Met; the FDR for protein identification and modification identification was both set as 1%.

The expression of proteins in each sample was determined as LFQ intensity (MaxLFQ), which is based on the accurate quantification of peptides in MaxQuant, as previously described by Cox and colleagues (Cox et al., 2014). Two to three independent proteomic assays were performed for each strain. Protein expression in the strains carrying mutated or deleted ClpX (CY706, CY708, CY1102, CY1179) was individually compared to that in strains carrying wild-type ClpX (CY598, CY1166). In total, eight comparisons were performed. Proteins that showed a p-value ≤ 0.1 and with a fold change higher than 2 in at least one comparison were defined as differentially expressed proteins (DEPs). Then, DEPs identified by proteome assay were compared to DEGs identified by RNAseq, and DEPs not being identified as DEGs by the RNAseq assay were regarded as the target of the ClpXP protease.

Bacterial cells from the early exponential phase (OD₆₀₀ of ∼0.4) were exposed to pH 2.5 for 5 min and plated on LB plates by series dilution. The survival rate was calculated by colony counting.