Bacteria slingshot more on soft surfaces

Adaptive responses greatly improve the competitive capacities of bacteria in diverse environments. Here, we investigate whether bacteria can adapt to a microenvironment with distinctive softness by examining the type-IV pili (TFP)-mediated motility of Pseudomonas aeruginosa cells on brush-like surfaces that are grafted with a layer of thermally sensitive polymer chains, where the softness of the brush-layer is tunable by applying a small temperature change (from 30 to 37 °C). We report that P. aeruginosa cells slingshot more on soft surfaces at a shear-thinning condition, which greatly facilitates their surface crawling by means of reducing energy dissipation. This adaptive response suggests that P. aeruginosa cells may be able to sense the local viscoelasticity and then deploy TFP to adapt to their physical surroundings.

O ver 99% of all bacteria live in biofilms that are multicellular surface-associated communities that appear ubiquitously in environmental, technological and medical settings [1][2][3] . The formation of gram-negative bacterial biofilms is facilitated by surface motility modes in the initial stages, where planktonic bacteria first become surface-associated and subsequently self-assemble into microcolonies 4,5 . Twitching motility, a well-known surface motility mode, mediated by type-IV pili (TFP) 6,7 , is employed by many bacterial species including the opportunistic pathogen Pseudomonas aeruginosa 8 , the gonorrheal bacterium Neisseria gonorrhoeae 9 and the predatory soil bacterium Myxococcus xanthus 10 . It has been known that twitching plays an essential role for these and other species of bacteria, enabling them to colonize abiotic/biotic surfaces and to form bacterial biofilms 6 that, on human tissues or medical settings, may lead to harmful or fatal infections 1,11 . Recently, a number of investigations 12,13 have shown that bacteria can use their TFP to walk or crawl on surfaces, in which walking or crawling motility can mediate cells to spread on surfaces with distinctive trajectories by means of short or long persistence lengths. Moreover, Jin et al. 14 have shown that bacteria can differently deploy their TFP to alternatively generate pull actions and slingshots in the crawling motility, where the velocities of slingshots are 20 times higher than those of pull actions. An intriguing question suggested by the aforementioned findings is whether bacteria can differentially deploy their TFP to generate proper motility types in response to different environmental cues.
Bacteria are thought to be able to adapt TFP-mediated motility in response to nutrient availability or chemical cues 15 because the growth 16 , extension and retraction of TFP are tightly regulated by chemotactic and cyclic-bis-(3 0 -5 0 )-dimeric-GMP (c-di-GMP) signalling pathways 8,17 . Aside from chemical cues, different abiotic/biotic surfaces often possess distinctive physical properties, such as viscoelasticity; for example, there is a shear modulus difference of at least six magnitudes between tissues and alloyed bone nail surfaces, yet bacterial biofilms can form on both 18 . However, it is less certain whether, especially at the singlecell scale, bacteria can adapt TFP-mediated motility or other surface motilities in response to their physical surroundings; only a few investigations suggest that bacteria do so 4,19,20 . Answering the question of whether bacteria can adapt TFP-mediated motility in response to specific physical cues, aside from answering a fundamental question in biophysics, has implications for ongoing research in seeking applicable biomaterials to prevent biofilm formation on medical settings. Here, we directly answer this question by investigating how individual P. aeruginosa cells deploy their TFP to crawl on a polymeric surface with a tunable softness. After carefully excluding other possible factors, we demonstrate that bacteria deploy their TFP to slingshot more on soft surfaces at a shear-thinning condition. Moreover, we find that this response facilitates cells to disperse on surfaces by means of reducing energy dissipation.

Results
Smart polymeric surfaces with a tunable softness. We grafted thermally sensitive poly(N-isopropylacrylamide) (PNIPAAM) chains on coverslips by surface-initiated atom transfer radical polymerization (SI-ATRP). SI-ATRP is a well-developed method that can be used to homogenously graft polymer chains on a surface with a high grafting density 21 . The synthetic methods are given in Methods and in Supplementary Fig. 1. It is well-known that adjusting the environmental temperature (T) around the transition point (T c C32°C) changes the conformation of PNIPAAM chains greatly on surfaces and in solutions 22,23 .
In the present study, raising T from lower (ToT c ) to higher (T4T c ) leads to the dehydration of brushy chains on the surface; in this transition, a fluid-like polymeric layer will collapse onto a surface to form a solid-like layer. By using a quartz crystal microbalance 22,24 , we confirmed that this transition does occur, resulting in a significantly increased hardness of surface (additional details are given in Supplementary Methods and in Supplementary Fig. 2). Moreover, the dehydration of PNIPAAM chains on surfaces can be monitored by their watercontacting angles (y w ) because this process increases surface hydrophobicity 25 . Supplementary Fig. 3f,g shows that dehydration occurs at the range of 31°CoTo36°C, which is consistent with existing literature 22 . Taken together, PNIPAAM brush surfaces can be tuned from soft to hard by only a small temperature change from 30 to 37°C.
Below the T c (T ¼ 30°C), fluid-like grafting chains can either stay at 'mushroom-like' or at 'brush-like' conformations on surfaces 26 , which is determined by their R g r 1=2 s , where R g is the radius of gyration of grafting chains at an unperturbed condition, r s is the number density of grafting chains at surface. Table 1 indicates that our prepared PNIPAAM surfaces possess a sufficiently high r s R g r 1=2 s 41 À Á , which enforces those grafting chains to stay at 'brush-like' conformations. As the static viscosity (Z) and the relaxation time (t) of polymer brushes depend on the number average molar mass (M n ) of grafting chains, we thereby prepared two distinctive PNIPAAM surfaces to control Z and t, on which the grafting chains possess distinctive M n , that is, M n ¼ 12,800 or M n ¼ 138,000; we denote them below as shortbrush or long-brush surfaces, respectively. Next, we estimated Z and t of short-brush or long-brush surfaces (see Methods and Table 1). We found that the major difference between short-and long-brush surfaces is their relaxation time, which is due to t / M 2 n . Finally, we examined whether these PNIPAAM brush surfaces are compatible to P. aeruginosa by investigating the growth of cells on long-brush surfaces at 30 and 37°C, respectively. Our results show normal bacterial growth and subsequent nascent biofilm formation on PNIPAAM brush surfaces, as shown in Supplementary Fig. 4, thus demonstrating that these brush surfaces are non-toxic for P. aeruginosa.
Tracking the crawling motility in P. aeruginosa. To isolate TFP-mediated motilities in P. aeruginosa 12,13 , we monitored the crawling of flagellum-deficient cells (DfliM) in various surface conditions using bright-field microscopy with high temporalspatial resolutions. Next, we analysed the motion of those cells by using a two-point tracking algorithm that can separately track the trajectories (r lead (t), r trail (t)) of leading (P lead ) and trailing poles (P trail ) for every individual cell over time 14 . We have previously shown that sub-pixel noises (0.5) inherently exist in those original trajectories 14 . To gain higher temporal-spatial resolutions, we    denoised those trajectories using a wavelet set of Daubechies that can significantly suppress the noises 27 , as shown in Fig. 1a. Thus, we can resolve instantaneous velocities v P lead t ð Þ j j ; v P trail t ð Þ j j ð Þ of P lead and P trail ranging from 0.01 to 3 mm s À 1 . As we have previously shown 14 , two distinctive features can be found in typical r lead (t), as shown in Fig. 1a, namely, (1) a continuous positioning of r lead (t), reflecting low-speed motions, the so-called pulling action and (2) a discontinuous positioning of r lead (t), denoting high-speed motions, the so-called slingshot with a much greater velocity than that of the pulling action ( Supplementary  Fig. 5). It is this greater velocity that enables us to detect the slingshot in r lead (t) by a threshold (0.09 mm s À 1 ). To further confirm that the pulling action and slingshot are mediated by TFP exclusively, we investigated the surface motilities of two additional mutants, DrhlADfliM and DpilADfliM. DrhlADfliM is deficient in producing rhamnolipids 28 and lacks flagellum; DpilADfliM lacks both the motility appendages of TFP and flagellum. Supplementary Fig. 5a clearly indicates that the pulling action and slingshot in the crawling motility are strictly mediated by TFP.
Slingshot on polymeric surfaces with distinctive softness. First, we investigated the crawling of cells on long-brush surfaces. We surprisingly observed that cells like to slingshot more on this surface at 30°C (Supplementary Movie 1) in comparison with those at 37°C (Supplementary Movie 2), as shown in Fig. 1b. We found the rate of slingshot (k s ) increases B3-fold by measuring multiple cells, as shown in Fig. 1c. Aside from the softness of surfaces, a number of possible factors may result in the increase of k s , including (1) a change of environmental temperature; (2) the chemical modification of the surface by PNIPAAM chains; (3) a change of surface hydrophobicity; (4) the alteration of the interaction between bacteria and surfaces and (5) the surface heterogeneity. To rule out the first possibility, we repeated the experiments on unmodified coverslips and the results showed that increasing T from 30 to 37°C does not alter k s , as shown in Fig. 1b,c. To address the second possibility, we investigated the crawling of cells on short-brush surfaces because chemically they are identical with long-brush surfaces. Our results for short-brush surfaces reveal that k s remains at the control level determined from unmodified coverslips at 30 or at 37°C (Fig. 1b,c). However, we also noticed that raising T from lower (ToT c ) to higher (T4T c ) increases hydrophobicity on short-brush surfaces (which it also does almost identically for long-brush surfaces, see Supplementary Fig. 3f,g). Therefore, no increase of k s on shortbrush surfaces can be used to exclude the third possibility.
To further address this point, we investigated the crawling of cells on hydrophobic glass surfaces. Supplementary Fig. 6 displays that k s remains at the control level. To address the fourth possibility, we evaluated the interaction between bacteria and surfaces using surface-shear stress assays 29 . Supplementary Fig. 7 shows that the interaction between bacteria and short-brush surfaces is similar to those of long-brush surfaces at corresponding temperatures of 30 and 37°C. As for the fifth possibility, we labelled long-brush surfaces using two different fluorescent dyes. Our results indicate that long-brush surfaces are homogenous by means of the spatial localization of grafting chains and the local hydrophobicity, as shown in Supplementary  Fig. 3a-c. After excluding all the aforementioned factors, we conclusively demonstrate that P. aeruginosa cells like to slingshot more if they meet a surface grafted with long polymer chains. (All additional and relevant details to the above points are given in Supplementary Information.) Trajectories on polymeric surfaces with distinctive softness. Figure 2a displays the typical trajectories of a crawling cell on a long-brush surface at 30°C and also at 37°C, where red and blue colours represent slingshots or pulling actions, respectively.
We noticed that when compared with the pulling action, the slingshot is more likely to change the direction of crawling. To quantify the directionality of crawling trajectories, we calculated the mean square displacements (MSDs) as a function of time intervals (Dt), and extracted their slope (k MSD ) by linear fits of log(MSD) as a function of log(Dt), that is, log(MSD)pk MSD log(Dt). k MSD ¼ 1.0 corresponds to random diffusive motion without a certain direction, whereas k MSD ¼ 2.0 corresponds to unidirectional ballistic motion. Figure 2b clearly indicates that raising T from 30 to 37°C leads to an increase of k MSD from 1.26 to 1.51 on long-brush surfaces, indicating that the crawling of the cell becomes more directional. To further confirm this result, we analysed multiple crawling trajectories on different surface conditions, as shown in Fig. 2c. Our results explicitly indicate that k MSD is smaller on long-brush surfaces at 30°C when compared with those that arise from control experiments ( Fig. 2c and Supplementary Fig. 8).

Discussion
Our results indicate that bacteria only slingshot more on longbrush surfaces than on short-brush surfaces at 30°C. The major difference in these surfaces is their relaxation time (t short ¼ 0.1 B1 s and t long ¼ 10B100 s), which prompts the question of why bacteria like to slingshot more on a soft surface that has a longer relaxation time. We estimated the shear rate of slingshot _ g s ð Þ on short-or long-brush surfaces using a relation _ g s ' 2v s =h eq , where v s is the averaged velocity of slingshots and h eq is the equilibrium height of the brush-layer. We found that t long _ g s ) 1, but t short _ g s $ 1, indicating that slingshot on long-brush surfaces can lead the brush-layer to tune into the condition of shearthinning. Therefore, one can expect that long-brush chains cannot relax during the t long after being sheared by a slingshot, which can markedly reduce the effective viscosity (Z eff ) following Z eff ¼ Zðt long _ g s Þ À 0:46 (refs 30,31). Note that the shear-thinning affects the low-speed pulling action in the duration of t long because the displacement of single pulling actions or slingshots is relatively smaller (Fig. 2a)  length. To account the overall shear-thinning effect, we calculated the time-average effective viscosity Z eff as where p s is the probability that the brush-layer stays at the shearthinning condition. One can expect that p s / e À t d =t , where t d is the time delay after one slingshot, as shown in Fig. 3a. The average value of t d can be estimated by B1/k s so that when tk s 4 41, equation (1) can be rewritten as A two-dimensional plotting of Z eff =Z as the function of t_ g s and tk s is shown in Fig. 3b, where colours represent the magnitude of Z eff =Z. It clearly indicates that at the regimes of t_ g s ) 1and tk s 4 41, Z eff is greatly reduced; for example, frequent slingshot (tk s 4 41) on the long-brush surface t_ g s ) 1 ð Þcan reduce energy dissipations up to 90%. Therefore, this TFP-mediated response can facilitate bacteria crawling on surfaces. Considering that P. aeruginosa continuously secretes extracellular polymeric substances (EPS) to form biofilms 32 , we also expect that cells will slingshot even more on a rigid surface if it is covered by secreted EPS, as both the viscosity and relaxation time of EPS 33 is comparable to those of long-brush surfaces.
In summary, we have conclusively demonstrated that P. aeruginosa cells slingshot more on a soft surface when t_ g s ) 1, which can greatly reduce surface viscosity and thus facilitate surface crawling. This adaptive response of P. aeruginosa suggests that those cells may be able to sense their local viscoelasticity and subsequently deploy their TFP to adapt to their surroundings. Figure 3c graphically summarizes our findings. In the real world, this type of adaptive bacteria behaviour may be observed on medical settings coated with a polymeric layer or even on human tissues. The deployment of TFP to slingshot can lead bacteria to spread and colonize on soft surfaces at a shear-thinning condition, hence facilitating biofilm formation. Therefore, this fundamental finding has implications for a number of research fields and may also provide an applicable approach for interfering with TFP-mediated motilities and any subsequent biofilm formation by controlling the surface softness of materials on which bacterial biofilms may form.
Preparation of PNIPAAM brush surfaces. The SI-ATRP was used to graft the polymer chains from surfaces 21 . First, the initiators were linked to the coverslips by refluxing CTMS in dry toluene at 125°C for 4 h, and then monomers (NIPAAM) were polymerized from initiator-modified coverslips in a DMSO solution containing free initiator (Methyl 4-(chloromethyl) benzoate), ligand (Me 6 TREN) and catalyst (cuprous chloride). This reaction was allowed to proceed at room temperature over 24 h and was terminated by inducting air. Note that the monomers were simultaneously polymerized in the solution. Thus, after the polymerization, the M n and the polydispersity indexes (PD.I.) of polymer chains in the solution are expected to be comparable to those chains grafted on the surface. Previous investigations have suggested that this estimation would lead an uncertainty of ±25% (refs 35,36). The above reactions were summarized in Supplementary Fig. 1a. The resultant coverslips were washed with fresh DMSO and sodium EDTA (aq.) and then dried with nitrogen before use. M n can be controlled by adjusting the mole ratio of monomer to initiator. The hydrophobic glass surfaces were prepared by coupling n-Dodecyltrimethoxysilane to coverslips in dry toluene.
Characterization of PNIPAAM brush surfaces. We first examined the PNI-PAAM brush surfaces using X-ray photoelectron spectroscopy (XPS) and found the nitric peak appearing in XPS spectrum, as shown in Supplementary Fig. 1b, suggesting that PNIPAAM chains were successfully grafted on the coverslips. By using gel permeation chromatography (GPC), we next estimated M n and PD.I. of the grafting chains by characterizing those free chains that were polymerized on the solutions. GPC results display that the PD.I. of those free polymer chains are o1.3, as shown in Supplementary Fig. 1c, which suggest that the molar mass distributions of those grafting chains are narrow. The dry thickness (h dry ) of the film formed by the PNIPAAM brush surfaces was characterized using an ellipsometry (M-2000V-ESM, J.A. Woollam Co.), in which translucent adhesive tapes (Scotch 810 Magic Tape, 3 M) were stuck on the back-surfaces to eliminate backside reflections. Afterwards, r s was determined by r s ¼ h dry r f N A /M n (ref. 37), where r f is the density of PNIPAAM dry films and N A is the Avogadro's number. Finally, we measured the contact angles (y w ) of a FAB droplet on short-or on long-brush surfaces at different temperatures, as is shown in Supplementary Fig. 3f,g. Our results clearly indicate that raising T from 31 to 36°C leads to the increasing of surface hydrophobicity on these surfaces, further demonstrating that PNIPAAM chains were successfully grafted on the coverslips.
Viscosities and relaxation times of PNIPAAM brush surfaces. In the regime of R g r 1=2 s 41, the equilibrium height (h eq ) of brush-layer can be calculated by h eq ¼ N p (2a 5 r s /3) 1/3 (ref. 26), where a ( ¼ 0.3 nm) is the monomer size 38 . Knowing the concentration (c ¼ r s M n /N A h eq ) of grafting chains in the brush layer, one can calculate Z using the intrinsic viscosities ([Z]) of grafting chains 39,40 . It has been known that the relaxation of grafting chains in polymer brushes is comparable to the relaxation of arms in star polymers 41,42 . Therefore, we estimated t of PNIPAAM brush surfaces using the relaxation time arising from star polymers that have comparable polymer concentrations and chain lengths. Note that relative errors in the estimation of t are around one magnitude. Additional details for the estimation of t are given in Supplementary Note 1.
Bacterial strains and flow cell experiments. Three mutants of P. aeruginosa (ATCC 15692), DfliM, DrhlADfliM and DpilADfliM, were used in this study, as is shown in Supplementary Table 1. Strains were grown on LB agar plates at 37°C for 24 h. Monoclonal colonies were inoculated into 5 ml polystyrene culturing tubes containing 1 ml of minimal medium (FAB) 43 and then grown in a shaker at 37°C, in which 30 mM glutamate was added as the carbon source. Bacterial cultures were harvested at exponential phase, as monitored by their optical densities at 600 nm (OD 600 E0.8). The resultant cultures were further diluted by adding 30 ml of the bacterial suspension into 1,000 ml of sterilized FAB. Flow cells (Denmark Technical University) used here were prepared and sterilized under the standard protocol. After an injection of bacterial cultures, the flow cell was typically given 15 min to allow for the cells to attach to the coverslip, any unattached cells were washed out afterwards. These surface-associated cells were further cultured either at 30.0±0.1°C or at 37.0±0.1°C by flowing the FAB with 0.6 mM glutamate with a constant flow rate (3.0 ml h À 1 ). Details for the construction of DrhlADfliM and DpilADfliM are given in Supplementary Methods.
Microscopy and tracking of crawling cells. An Olympus IX81 inverted-microscope equipped with a Â 100 oil objective and an sCMOS camera (Neo, Andor) was used to acquire the bright-field images (528 Â 512 pixels) of surface-associated cells with a frame rate of 10 fps. A typical movie contains B18,000 images. Seventy-five movies were acquired and analysed in total, in which the 16-bit greyscale images were first converted to binary images for the detection of bacteria with a standard image processing algorithm, and the x-y positions of leading and trailing poles in single cells were second determined and linked individually over time by using our established two-point tracking algorithm 14 . The resultant trajectories of leading or trailing poles were further denoised using Daubechies wavelets (Fig. 1a) 27 . A velocity threshold (0.09 mm s À 1 ) was used to detect the slingshot from the time series of instantaneous velocity.