Spatial Regulators for Bacterial Cell Division Self-Organize into Surface Waves in Vitro

See allHide authors and affiliations

Science  09 May 2008:
Vol. 320, Issue 5877, pp. 789-792
DOI: 10.1126/science.1154413


In the bacterium Escherichia coli, the Min proteins oscillate between the cell poles to select the cell center as division site. This dynamic pattern has been proposed to arise by self-organization of these proteins, and several models have suggested a reaction-diffusion type mechanism. Here, we found that the Min proteins spontaneously formed planar surface waves on a flat membrane in vitro. The formation and maintenance of these patterns, which extended for hundreds of micrometers, required adenosine 5′-triphosphate (ATP), and they persisted for hours. We present a reaction-diffusion model of the MinD and MinE dynamics that accounts for our experimental observations and also captures the in vivo oscillations.

How macroscopic order can emerge from local molecular interactions is a general problem of cell biology. Organization can emerge from a limited number of constituents that diffuse and locally interact with each other, provided that the system is out of equilibrium (1). A reaction-diffusion mechanism of pattern formation has been observed in a number of systems (2, 3) and has also been suggested to lead to spatial protein organization in vivo (47).

The oscillations of the Min system in Escherichia coli are a strong candidate for a reaction-diffusion system in vivo (6). This system consists of the proteins MinC, MinD, and MinE (8), which oscillate between the poles of the rod-shaped bacterium (9, 10) and thereby select the cell center as the site for division septum formation (11). The Min proteins are crucial for accurate cell division. Mutants lacking the Min system are prone to divide asymmetrically, which gives rise to DNA-free minicells. MinD is an adenosine triphosphatase (ATPase) that dimerizes in the presence of adenosine 5′-triphosphate (ATP) and binds to the lipid membrane via amphipathic helices (12). In the cell, MinD assembles on the cytoplasmic membrane covering roughly half of the cell. MinE binds to membrane-bound MinD and induces ATP hydrolysis by MinD (13). Subsequently, both proteins detach from the membrane, and MinD reassembles in the opposite half of the cell. MinE is predominantly located at the rim of the MinD domain and forms a ring close to the cell center (14). MinC follows the movement of MinD directly and is not required for the oscillations (9, 10). However, it prevents the assembly of FtsZ, which is known to be the first protein to localize at the future site of cell division (15). By inhibiting FtsZ assembly at the cell poles, the Min system restricts the formation of the division septum to the cell center.

Various theories have been suggested to explain the Min oscillations (6). Some models propose that no spatial markers are required to generate this pattern (6) and that spontaneous oscillations will form even from a uniform initial state, a phenomenon called dynamic instability. Thus, Min-protein self-organization should not be restricted to living bacteria, but could also be tested in vitro (16). We designed an experimental approach with a minimal number of components to systematically explore the Min system.

Our in vitro system was based on a supported lipid bilayer to mimic the cell membrane, which was immersed in buffer containing the fluorescently labeled proteins MinE (Alexa647) and MinD (Bodipy-FL) (17). In the presence of ATP, MinD attached to the membrane, forming a dense, homogeneous protein layer. When we added MinE to the buffer, planar surface waves formed on the membrane within 1 hour (Fig. 1 and movie S1). These waves moved in a distinct direction across the membrane and persisted for several hours. They were composed of spatially periodic bands of MinD and MinE, separated by troughs devoid of protein. The fluorescence intensities indicated a characteristic protein-density distribution parallel to the propagation direction (Fig. 2, A and B, and movie S2). From the leading edge of the wave, the density of MinD increased and decreased sharply toward the trailing edge. The MinE density displayed a similar intensity profile, but with a shallow linear rise in fluorescence and a sharp decrease at the trailing edge. In a propagating wave, the density maximum of MinE followed the maximum of MinD and formed a sharp line along the trailing edge of the wave. This pattern resembles the situation in the cell, where a dynamic MinE ring moves toward the pole, detaching MinD from the cell membrane (14).

Fig. 1.

Min-protein surface waves in vitro. Confocal images of self-organized protein waves on a supported lipid membrane, MinD (1 μM), doped with 20% Bodipy-labeled MinD (green), MinE (1 μM), doped with 10% Alexa647-labeled MinE (red). Scale bar, 50 μm.

Fig. 2.

Quantitative characterization of protein waves. (A) Confocal image of Min protein waves on a lipid membrane. MinD (1 μM), green channel; MinE (1.5 μM), red channel. Scale bar, 50 μm. (B) Intensity profile plots for MinD and MinE waves along the rectangular area shown in (A). Arrows in (A) and (B) indicate direction of wave propagation. (C) Kymographs for MinD and MinE along the rectangular area shown in (A). Surface wave velocity (D) and wavelength (E) as a function of MinE concentration (MinD at 1 μM). Data of the velocities fitted to Hill equation yielded Vmax = 0.94 μm/s (left). Each data point has been obtained from n = 3 independent measurements. Error bars in (D) and (E) indicate SD.

The dynamics of the waves were influenced by the concentration of MinE. For given concentrations of MinD and MinE, waves moved at constant velocity and wavelength (Fig. 2C). When we increased the MinE concentration from 0.5 to 5 μM, the average propagation velocity changed from 0.28 μm/s to a value of 0.8 μm/s (Fig. 2D). At the same time, the average wavelength decreased from 100 to 55 μm (Fig. 2E). At concentrations below 0.2 μM, waves still formed erratically, but the system did not settle into a well-defined state (see movie S3). Wave formation crucially depended on the hydrolysis of ATP. Consistent with previous investigations (13), MinD did not attach to the membrane in the absence of ATP, and wave formation could not be observed. When we supplied the nonhydrolyzable ATP analog, adenosine 5′-O-(3-thiotriphosphate) (ATP-γ-S), MinD formed a homogeneous protein layer on the membrane. MinE attached to membrane-bound MinD, but was not able to induce wave formation (fig. S6). These findings confirmed that surface waves were only formed when energy was dissipated.

Without MinE, MinD was homogenously distributed on the membrane. How did the presence of MinE lead to planar surface waves? One to 5 min after addition of MinE, the dense layer of MinD on the membrane became unstable and depletion zones appeared, which coincided with regions of high MinE density (Fig. 3A, at 4 min, and movie S4). These areas without membrane-bound MinD increased in size at constant MinE concentration in the buffer and were delimited by a continuous line of MinE. Depletion zones merged as they grew and encountered each other. MinE borders coalesced, leaving only MinD islands, which eventually disappeared (Fig. 3A, at 6 min). Subsequently, MinD reattached to the protein-free parts of the membrane with MinE distributed inhomogeneously around the newly formed MinD layer (Fig. 3A, at 10 min). Discrete protein-free ripples moved independently from each other across the membrane. The direction of motion of a ripple was governed by the asymmetry of MinE distribution, with MinE mainly located as a sharp line at the leading edge (see Fig. 3A and movies S4 and S5). When two ripples collided they fused and formed a larger protein-free area, which then moved in one direction. After about 1 hour all discrete ripples had synchronized into a highly regular pattern of parallel moving protein bands (Fig. 3A, at 18 and 38 min, see movie S5). In some instances, we observed rotating spirals of Min proteins (Fig. 3, B and C, and movies S6 to S8). The wavelengths and velocities of the waves emanating from a spiral did not vary with the distance from the spiral center and were the same as for the planar waves. When waves from two neighboring counter-rotating spirals collided, they fused to form parallel protein bands (Fig. 3C and movie S9). This might be the reason why we most frequently observed parallel waves.

Fig. 3.

Initiation of protein surface waves. (A) Starting from a homogenous distribution of MinD (1.5 min after addition of MinE), addition of MinE led to dynamic instability (4 to 38 min): First, MinD detached from the membrane, and after reattachment, protein-free ripples in the protein layer eventually synchronized to a regular pattern of parallel surface waves. MinD (1 μM), green channel; MinE (1.5 μM), red channel. (B) Spiral waves formed by Min proteins. (Left) Only labeled MinE is shown (MinD, 1 μM; MinE, 1 μM); (right) labeled MinD and MinE are shown (MinD, 1 μM; MinE, 1 μM), (C) Double spirals formed by Min proteins; only labeled MinE is shown (MinD, 1 μM; MinE, 0.5 μM). The star marks the center of the double spiral. Scale bar, 50 μm if not differently indicated (movies in SOM).

To further investigate the mobility of the proteins during wave propagation, we performed fluorescence photobleaching experiments. The bleached area of MinE or MinD remained at its original position on the membrane, while the wave was propagating (figs. S9 and S10). This implies that the waves were not the result of protein translocation in the plane of the membrane, but were generated by sequential rounds of detachment and reattachment of proteins from the soluble pool.

The waves we observed experimentally were qualitatively different from the behavior predicted by existing theories (6) (see movies S12 and S13). These theories are either based on a classical reaction-diffusion mechanism (18, 19), where cooperative MinD binding to the membrane is essential to generate oscillations, or they assume attractive interactions between MinD molecules bound to the membrane (20). From the observation that regions of high MinE densities initiated MinD detachment and eventually pattern formation (Fig. 3A), we developed a computational model [see supporting online material (SOM) text] that includes cooperative effects during MinE binding to the membrane, similar to those suggested in (18, 21). MinE cooperativity proved to be crucial to reproduce the dynamic patterns observed experimentally (Fig. 4 and movies S10 to S11). Our theory also captured a similar protein-density distribution within a protein band (Fig. 4B), as well as the dependencies of the wavelengths and velocities on the MinE concentration (Fig. 4D).

Fig. 4.

Computational model of Min-protein dynamics. (A) Planar surface wave solutions of the dynamic equations (see SOM). (B) Profile plots and kymographs for MinD and MinE waves along the rectangular area shown in (A). Arrows indicate direction of wave propagation. Maxima of the MinE distribution follow maxima in the MinD distribution. (C) Spiral solution for the same parameter values as in (A). (D) Dependence of the velocity and wavelength of the surface waves on the MinE concentration. Whereas the velocity increases markedly with increasing MinE concentrations, the wavelength shows a moderate decrease. In all simulations, periodic boundary conditions have been used. The size of the domains in the simulations were 900 μm × 900 μm, extracts are shown in (A) and (B). Scale bar, 100 μm. Error bars in (D) and (E) indicate SD. Parameter values are given in the SOM text.

Several features of the in vitro structures of the Min system are strongly reminiscent of the Min oscillations in vivo: MinD was distributed homogenously on the membrane in the absence of MinE, whereas dynamic patterns could be observed only in the presence of MinE; MinE was found predominantly localized at the trailing edge of a moving MinD band (14, 22). We characterized the velocity and wavelength of the surface waves as a function of MinE concentration and conclude that the frequency of the oscillations increases with an increasing MinE/MinD ratio (14). An explanation for the different length scales of the patterns observed in vitro and in vivo is given by our theoretical description: When we used lower values for the diffusion constants of the membrane-bound proteins in our model, we could also reproduce the Min oscillations observed in the cell (23) (see SOM text). One possible reason for lower diffusion constants in vivo could be molecular crowding in the cytoplasmic membrane of E. coli. Thus, the mechanism generating the surface waves in vitro may also drive the Min oscillations in vivo. By extending the experiments to the rod-shaped geometry of an E. coli cell, it could be possible to reconstitute pole-to-pole oscillations as observed in vivo or, eventually, even the whole prokaryotic cell division machinery. Here, we have shown that complex biological behavior can emerge from a limited number of components, namely, two proteins, a membrane, and ATP.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S13

Table S1

References and Notes

Movies S1 to S13

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article