Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Phys Rev Lett. Author manuscript; available in PMC 2017 July 26.
Published in final edited form as:
PMCID: PMC5529051

Barrier-crossing in Escherichia coli chemotaxis


Cells live in time varying environments with multiple desirable locations separated by unfavorable regions. To study cell navigation in spatiotemporally varying environments, we developed a microfluidic “race-track” device to create traveling attractant waves with multiple peaks and a tunable wave speed (υw). We found that while the population-averaged chemotaxis drift speed (υd) increases with υw for low υw, it decreases sharply for high υw. Our single cell measurements revealed that this reversed dependence of υd on υw is caused by a “barrier-crossing” phenomenon, where a cell hops backwards from one peak attractant location to the peak behind by crossing an unfavorable (barrier) region with low attractant concentrations. The barrier-crossing process is enabled by the cell’s random motion, which acts as temperature in thermally activated processes. Our simulation results and theoretical analysis showed that the backward barrier is lowered by υw and the backward drift speed depends exponentially on υw, leading to the observed sharp drop in υd for high υw. The barrier-crossing effect is further confirmed in double well experiments.

Sensing and responding to changes in external environments are critical for the survival of living organisms. One of the well-studied model systems is bacterial chemotaxis. Bacteria use their transmembrane chemoreceptors to sense their environments and control their motion in search of places with more favorable conditions [13]. In a homogeneous environment, an E. coli cell performs the run-and-tumble random walk allowing it to explore its environment [4]. In the presence of an attractant gradient, E. coli cells bias their random walk towards the preferred direction by lengthening the run time in the “correct” direction. The intracellular biochemical circuit that allows an E. coli cell to compute gradient has been studied extensively in the past decades [59]. Predictive models have been developed based on knowledge of the bacterial signaling pathway and quantitative molecular and celluar experiments [1013]. A modeling framework based on the intracellular signaling dynamics and the motor response has also been developed to study cellular and population behaviors [1417].

Cells live in complex environments together with other cells. There can be multiple favorable locations separated by unfavorable regions. Cells can also emit chemical signals and act as moving sources of attractants [18]. Can a cell find its way out of a local optimum location to explore the environment globally? Can a cell track a moving attractant source? Here, we investigate these questions by developing a microuidic device to create environments with multiple attractant peaks that move with a tunable speed. In particular, we created a traveling wave of attractant concentration in an annulus (race-track) channel as shown in Fig.1. Both population level behaviors and individual cell trajectories were measured for traveling wave attractant profiles with different wave speeds (υw). Our measurements showed that the population-averaged chemotaxis drift speed (υd) increases with υw for small υw. Surprisingly, we observed a critical wave Surprisingly, we observed a critical wave speed, beyond which υd decreases sharply with υw instead of reaching a saturating value. Our individual cell trajectory data revealed that cells can hop from one peak attractant position to another by crossing a barrier region with lower attractant concentrations, and the backward hopping probability sharply increases with υw.

FIG. 1
Experiment setup. (a) The panorama of PDMS chip. (b) The zoomed-in picture of the observation channel. The circumference of the observation channel is λ = 800 µm. (c) The spatiotemporal profile of attractant (MeAsp) concentration in the ...

To explain the experimental observations quantitatively, we studied a theoretical model of chemotaxis motion based on the intracellular signaling dynamics. Our model analysis showed that E. coli chemotactic behavior can be mapped to a thermally activated motion in an effective energy landscape, with the cells random motion acting as the source of thermal fluctuation and an effective potential determined by the ligand concentration profile. The effective potential barrier height for backward hopping is lowered by υw, which results to a backward drift speed that depends exponentially on υw. This exponential backward drift speed leads to the sharp drop in υd at high υw. Finally, the barrier-crossing phenomenon is confirmed by a “double well” experiment with different barrier heights..

We first describe our microfluidic device shown in Fig.1(a). The device is composed of concentration modulating parts, an annular observation channel, two agarose adding channels, and a cell loading channel. Attractant and buffer solutions, after being well mixed in the modulating parts, are pumped into the modulating parts with time varying injection speed. Details of the observation channel are presented in Fig.1(b). The connection channels between observation channel and source channels are filled with agarose gel and serve as control points. The hydrogel added from the agarose inlets is used to prevent bacteria escaping and avoid net flow that affects bacterial motility. The attractant concentration profile in the observation channel is determined by diffusion through the four control points. The same oscillation of attractant concentration (amplitude and period) was introduced in the four source channels with a phase delay of π/2 between two adjacent control points. As a result, the attractant molecules diffuse into the annular channel, forming a traveling-wave concentration field. The wave speed υw is determined by the driving attractant period at the four control points. We test the attractant concentration in the observation channel by adding fluorescein in the attractant solution and the spatiotemporal concentration profile is shown in Fig.1(c) for υw = 1 µm/s. The cell loading channel and observation channel are linked by a narrow pass. Because the attractant concentration in the observation channel is always higher than that in the loading channel during experiments, cells that are loaded to the cell loading channel can chemotax to the observation channel through the narrow pass and seldom escape out of it. [see supplementary material (SM) for detail of chip layout and fabrication].

Cell motion in the observation channel were imaged by using dark field lens. E.coli (wild type RP437) swimming in the traveling attractant wave (α-methyl-DL-aspartate (MeAsp)) with different wave speeds (υw) 0.67 µm/s, 1 µm/s, 2 µm/s, 4 µm/s, 8 µm/s, 10 µm/s, 13.3 µm/s, are tracked and analyzed [see SM for details]. The spatiotemporal cell density profiles for different wave speeds are shown in Figs.2(a)–(c). The bacterial chemotactic behavior depends strongly on υw. For υwµm/s, most bacteria form a cluster following the crest of the attractant wave [Fig.2 (a)–(b)]. Occasionally, a cell escapes from the cluster and moves in a backward direction opposite to the attractant wave as indicated by the red arrow in Fig.2(b). However, for υw > 4µm/s, such backward motion becomes more frequent leading to a more diffused cell distribution [Fig.2 (c)].

FIG. 2
Dynamics of bacterial population. The spatiotemporal cell density profiles for different wave speeds 1 µm/s (a), 4 µm/s (b), and 10 µm/s (c). The normalized cell density and attractant concentration is represented by the color ...

To characterize the bacterial population dynamics at different υw quantitatively, we calculated the bacterial drift velocity υd by averaging the velocities of all cell trajectories within a period [see SM for details for trajectory extraction]. As shown in Fig. 2(d), for small wave speed υw ≤ 2µm/s, we have υdυw as cells can follow the attractant wave. For 8µm/s > υw > 2µm/s, υd starts to deviate from υw but still increases with υw albeit sub-linearly. This slowing down is likely caused by the effect of a finite adaptation time of E. coli in tracking/computing the attractant gradient [15, 16, 19]. However, the most surprising observation is that υd decreases sharply with υw instead of saturating to a constant value when υw ≥ 8 µm/s. In the rest of this paper, we try to understand the observed non-monotonic dependence of υd on υw, in particular the sharp decrease in υd for large υw.

To characterize the relative motion of individual cells with respect to the traveling wave, we define the phase shift of a cell for a given period as Δ[var phi]T [equivalent] 2πΔx/(υwT), where Δx is the net displacement along the direction of the attractant wave in a period T. If a cell follows the wave exactly, we have Δ[var phi]T = 2π. If a cell hops backwards to the peak behind the current one during time T, we have zero net displacement Δx = 0, and Δ[var phi]T = 0. Two representative trajectories for Δ[var phi]T ≈ 2π and 0 are shown in Fig.3(a). About 50 cells for the wave speeds of 2 µm/s and 8 µm/s are analyzed by manually tracking their trajectories over a complete period. Fig.3(b) shows the probability distributions of Δ[var phi]T for υw = 2 µm/s and υw = 8 µm/s. For υw = 2 µm/s, the Δ[var phi]T distribution peaks around a large Δ[var phi]T ~ 1.5π. However, for υw = 8 µm/s, an additional peak appears in the Δ[var phi]T distribution near Δ[var phi]T ~ 0, indicating the significance of backward hopping, which is responsible for the significant reduction in υd for large υw.

FIG. 3
Single cell dynamics. (a) Two representative single-cell trajectories with Δ[var phi]T ≈ 2π (left panel) and Δ[var phi]T ≈ 0 (right panel) from experiment with υw = 8 µm/s. The arrow shows the direction ...

To understand both the population-level and individual cell behaviors quantitatively, we used the signaling pathway-based E. coli chemotaxis simulator (SPECS) [19] instead of the filter-function based models [2022]. The advantage of SPECS is that it incorporates the internal signaling pathway dynamics with the movements of individual cells (see SM for details of SPECS simulation). The dependence of υd on υw from SPECS agrees with our experimental data [Fig.2(d)]. We also studied statistics of Δ[var phi]T of individual cells for different traveling attractant wave speeds in SPECS. As shown in Fig.4, the Δ[var phi]T distributions exhibit multimodality with peaks centered around 2π and 0, and the proportions of bacteria distributed around different peaks change significant with υw. For υw = 2 µm/s, a large proportion of cells have Δ[var phi]T around 2π and only about < 20% of the population distributes near 0. As υw increases, the proportion of the cells with Δ[var phi]T ~ 0 increases, and eventually dominates at high υw.

FIG. 4
The statistic of Δ[var phi]T for different wave speed from SPECS. The multimodality of density probability is caused by the “barrier-crossing” between the neighboring local well. The effective potentials for υw = 0 and ...

In our previous work, a mean field theory based on intracellular signaling dynamics was developed for studying population level bacterial chemotaxis behaviors [15, 17]. Briefly, the tumbling rate zt = τ−1(a/a0)H is modulated by chemoreceptors activity a, where τ and a0 are the average run time and activity of chemoreceptors at steady state, H (≈ 10) is the Hill coefficient [23]. The total frequency of a cell changing its direction is the sum of the rotational diffusion coefficient (z0) and the tumbling rate (zt): z(a) = zt + z0. Thus, the average run time is [z macron]−1z−1|(a=ā) with ā the average activity of all cells at position x, and the average run distance is υ[z macron]−1 with υ the run speed. The dynamics of the receptor activity is governed by the local attractant concentration and the receptor methylation level, which has a slow dynamics and essentially carries a memory of the cells environment in the past. Therefore, when cells move in a chemical gradient, the average activity of the left-moving cells at position x is different from the right-moving cells at the same position, as these two populations carry different receptor methylation levels. The activity difference leads to a difference in the tumbling frequency Δz, which eventually drives the bacterial chemotactic drift.

In our experiments with traveling attractant waves, it is convenient to study the system in a co-moving frame with the attractant by using the transformation x′ = xυwt. The dynamics of the bacterial density ρ(x′, t) in the moving frame is given by:


which describes bacterial chemotaxis motility as the thermal motion of particles moving in an external potential [24]. The attractant field gives rise to the external potential and the bacterial random walk acts as the thermal fluctuation. As shown in details in SM, the diffusion coefficent D(x′) and the effective potential U(x′) can be expressed as:



The effective potential U(x′) given above by Eq. 2 depends on the wave speed υw. The potential contributed by the attractant profile alone (i.e., υw = 0) is periodic (black curve in Fig.4 inset). The effect of υw is to tilt this periodic potential along the wave direction (x) to form a tilted washboard potential for υw > 0 (red curve in Fig.4 inset). This can be seen by approximating D as a constant in Eq. 2, which leads to a term υwx′/D for the potential U(x′). In this washboard potential, a cell quickly moves to its closest well and stay there until the random walk motion drives it over a barrier into a neighboring well. Based on the classical Kramers theory of energy barrier crossing [25], the hopping rates along (forward) or against (backward) the attractant wave direction (rf or rb) depends on the barrier height exponentially:


where r0 is the base attempt rate, and α is the hopping rate for stationary wave. The constants β1, β2 are independent of υw (see SM for detailed derivations). The discrete hopping events from one well to another explain the multimodality of the probability distribution of Δ[var phi]T [Fig. 3(b) and Fig.4]. As υw increases, the backward hopping rate rb increases and the percentage of cells with Δ[var phi]T = 0 increases. Put together, υd is given by:


where λ = 800 µm is the peak-to-peak distance of the traveling wave. For small υw, the hopping rates are small, so υdυw. However, due to the exponential dependence of rb on υw, there exists a critical wave speed (υc) where d/dυw = 0, and the drift velocity is dominated by the backward hopping term rb and decreases sharply for υw > υc. In our experiments, the attractant wave amplitude had a weak dependence on υw, which leads to α = α0(υw)d, where d ≈ 1.34 is determined from experiments, and α0 is a υw-independent constant (see SM for details). Quantitatively, Eq. 4 fits the experiments and SPECS simulations well with α0 = 0.0141/s, β1 = 0.0713 s/µm and β2 = 0.4915 s/µm as shown in Fig.2(d) (Eq. 4 is replaced by the full Eq. S26 to predict υd at υw = 13.3µm/s, see SM for details).

To verify the barrier crossing effect with better statistics, we studied chemotaxis in a “double well potential”. The double well potential is achieved in the same device by giving a static high concentration at control points P1 and P3; and a static low concentration at P2 and P4. Cells were concentrated initially near P1 by lowering the concentration at P3. Once a majority cell population is established around P1, the concentration at P3 is raised to form the double well potential [Fig.5(a)]. The cells hop probabilistically between P1 and P3, causing cell density in P1 to decay exponentially to 0.5 over time [Fig.5(b)]. When we increase the potential barrier by decreasing the concentration at P2 and P4, the decay rate decreases as expected from the transition state theory [Fig.5(b) inset].

FIG. 5
Bacterial motion in a “double well potential”. (a) The concentration profile for generating a stationary “double well potential”. By controlling the attractant concentration difference between P1(P3) and P2(P4), we can ...

In summary, by developing a “race-track201D; microuidic device we investigate bacterial chemotaxis behaviors in response to traveling attractant waves. The underlying mechanism for the observed non-monotonic dependence of the chemotaxis drfit speed on the attractant wavespeed is understood by using a computational model based on realistic signaling dynamics and cell motility. The results on single cell response to spatio-temporal signals as well as the integrated approach combining quantitative microuidics experiments with biology-based models are critical in studying more complex biological phenomena, such as collective behaviors due to cell-cell communication through biochemical signaling pathways [2628].


This work is partially supported by the National Natural Science Foundation of China (11434001 and 11674010). The work by YT is supported by NIH (GM081747).


1. Berg HC. Physics Today. 2005;5:53.
2. Kentner D, Sourjik V. Current opinion in microbiology. 2006;9:619. [PubMed]
3. Porter SL, Wadhams GH, Armitage JP. Nature Reviews Microbiology. 2011;9:153. [PubMed]
4. Berg HC, Brown DA, et al. Nature. 1972;239:500. [PubMed]
5. Alon U, Surette MG, Barkai N, Leibler S. Nature. 1999;397:168. [PubMed]
6. Sourjik V, Berg HC. Proceedings of the National Academy of Sciences. 2002;99:123. [PubMed]
7. Shimizu TS, Tu Y, Berg HC. Molecular systems biology. 2010;6:382. [PMC free article] [PubMed]
8. Yuan J, Branch RW, Hosu BG, Berg HC. Nature. 2012;484:233. [PMC free article] [PubMed]
9. Sourjik V. Trends in microbiology. 2004;12:569. [PubMed]
10. Tu Y. Annual review of biophysics. 2013;42:337. [PMC free article] [PubMed]
11. Yi T-M, Huang Y, Simon MI, Doyle J. Proceedings of the National Academy of Sciences. 2000;97:4649. [PubMed]
12. Tu Y, Shimizu TS, Berg HC. Proceedings of the National Academy of Sciences. 2008;105:14855. [PubMed]
13. Keymer JE, Endres RG, Skoge M, Meir Y, Wingreen NS. Proceedings of the National Academy of Sciences of the United States of America. 2006;103:1786. [PubMed]
14. Erban R, Othmer HG. SIAM Journal on Applied Mathematics. 2004;65:361.
15. Si G, Wu T, Ouyang Q, Tu Y. Physical review letters. 2012;109:048101. [PMC free article] [PubMed]
16. Zhu X, Si G, Deng N, Ouyang Q, Wu T, He Z, Jiang L, Luo C, Tu Y. Physical review letters. 2012;108:128101. [PMC free article] [PubMed]
17. Hu B, Tu Y. PLoS Comput Biol. 2014;10:e1003672. [PMC free article] [PubMed]
18. Budrene EO, Berg HC. 1991 [PubMed]
19. Jiang L, Ouyang Q, Tu Y. PLoS Comput Biol. 2010;6:e1000735. [PMC free article] [PubMed]
20. De Gennes P-G. European Biophysics Journal. 2004;33:691. [PubMed]
21. Clark DA, Grant LC. Proceedings of the National Academy of Sciences of the United States of America. 2005;102:9150. [PubMed]
22. Celani A, Vergassola M. Proceedings of the National Academy of Sciences. 2010;107:1391. [PubMed]
23. Cluzel P, Surette M, Leibler S. Science (New York NY) 2000;287:1652. [PubMed]
24. Van Kampen NG. Stochastic processes in physics and chemistry. Vol. 1. Elsevier; 1992.
25. Kramers HA. Physica. 1940;7:284.
26. Park S, Wolanin PM, Yuzbashyan EA, Silberzan P, Stock JB, Austin RH. Science. 2003;301:188. [PubMed]
27. Nakagaki T, Yamada H, Tóth Á. Nature. 2000;407:470. [PubMed]
28. Hol FJ, Dekker C. Science. 2014;346:1251821. [PubMed]