Self-Assembly of MinE on the Membrane Underlies Formation of the MinE Ring to Sustain Function of the Escherichia coli Min System*

Background: MinE of the Min system forms a ring-like structure that plays a critical role in triggering the oscillation cycle. Results: MinE self-assembles into fibrillar structures on the membrane through the N-terminal domain. Conclusion: A self-assembly mechanism may underlie the formation of the MinE ring. Significance: This study suggests that self-assembly of MinE on the membrane is a fundamental property of the Min system. The pole-to-pole oscillation of the Min proteins in Escherichia coli results in the inhibition of aberrant polar division, thus facilitating placement of the division septum at the midcell. MinE of the Min system forms a ring-like structure that plays a critical role in triggering the oscillation cycle. However, the mechanism underlying the formation of the MinE ring remains unclear. This study demonstrates that MinE self-assembles into fibrillar structures on the supported lipid bilayer. The MinD-interacting domain of MinE shows amyloidogenic properties, providing a possible mechanism for self-assembly of MinE. Supporting the idea, mutations in residues Ile-24 and Ile-25 of the MinD-interacting domain affect fibril formation, membrane binding ability of MinE and MinD, and subcellular localization of three Min proteins. Additional mutations in residues Ile-72 and Ile-74 suggest a role of the C-terminal domain of MinE in regulating the folding propensity of the MinD-interacting domain for different molecular interactions. The study suggests a self-assembly mechanism that may underlie the ring-like structure formed by MinE-GFP observed in vivo.

Binary fission in bacteria allows cells to divide from the center of the long axis of the cell, which leads to their physical separation into two daughter cells. The Min system, comprising the three proteins, MinC, MinD, and MinE, is one of the mechanisms that regulates the spatial precision of division site placement in Escherichia coli (1). MinD associates with the membrane, and its distribution is restricted to one-half of the cell by MinE (2). The coordinated activity of MinD and MinE leads to the formation of the MinD polar zone (2). In addition, MinE forms a ring-like structure, the MinE ring, at the medial edge of the MinD polar zone that triggers the pole-to-pole oscillation of both subcellular structures (3,4). This "MinDE oscillator" partitions the cell division inhibitor, MinC, toward the polar sites to block aberrant polar divisions through the direct interaction of MinC and MinD (5)(6)(7)(8).
The Min system functions through complex protein-protein and protein-membrane interactions to maintain the subcellular localization and oscillation of the Min proteins. In brief, during assembly of the MinD polar zone, the membrane interaction (9,10) and the cooperative assembly of MinD on the membrane (11,12) are predominant activities that allow growth of the MinD polar zone toward the midcell. MinC is recruited to the polar zone by MinD and becomes active to block division at the polar sites. When growth of the MinD polar zone approaches the midcell, the MinE ring starts to assemble. MinE targets to the membrane by the recruitment of MinD and by direct membrane interaction (3,13). The mechanism underlying the assembly and disassembly of the MinE ring is not known. The presence of the MinE ring at the midcell not only arrests further growth but also initiates disassembly of the MinD polar zone (4,14). Disassembly of the MinD polar zone involves stimulation of MinD ATPase activity by MinE, which results in release of MinD from the membrane and recycle to the other end of the cell for assembly of the polar zone at the opposite pole (11,15,16). At the end of the polar zone disassembly, the MinE ring collapses, and the protein molecules redistribute to the midcell for the formation of a new ring.
MinE acts as a molecular switch in driving the oscillation from one pole to the other. It is a small protein that can be dissected into three functional regions: the N-terminal membrane-targeting motif (MinE [2][3][4][5][6][7][8][9][10][11][12], the MinD-interacting domain (MinE [13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31], and the C-terminal dimerization domain (MinE  ). Recent evidence suggests that MinE exhibits dynamic conformational flexibility upon association with different interacting partners. The membrane-targeting motif folds into an amphipathic helix after its association with the membrane (17). The MinD-interacting domain folds into an ␣-helical conformation when MinE forms a protein complex with MinD (18). The C-terminal domain of MinE forms homodimers in solution carrying one ␣-helix and two ␤-strands in each monomer arranged in an anti-parallel fashion, as reported in an earlier NMR study (19). However, the MinD-interacting domains of MinE from both Helicobacter pylori and Neisseria gonorrhoeae exist as a ␤-strand that is sequestered into the dimer interface of the C-terminal domain (20,21). These studies suggest that drastic conformational transitions of MinE may occur when MinE encounters different interacting partners. Nonetheless, the correlation between the MinE folds and the MinE ring remains unclear.
In this work we investigated whether self-assembly of MinE underlies the formation of the MinE ring. We discovered that MinE formed fibrillar structures on supported lipid bilayers (SLBs), 6 a defined in vitro system, to reveal the protein function. The N-terminal domain of MinE (MinE 1-31 ) is likely involved in the formation of the membrane-associated fibrils based on the following evidence. MinE 1-31 formed fibrillar structures in solution that share common morphological features and dyestaining properties to amyloid fibrils. We confirmed that a segment (residues 19 -27) within the MinD-interacting domain, which forms an anti-parallel ␤-sheet and is sequestered in the dimer interface of MinE (20,21), was responsible for the amyloidogenic function. The formation of amyloid fibril from a protein segment often reflects the self-associating property of the intact protein under native conditions (22)(23)(24)(25). Studies of the MinE mutants I24N and I25R supported the involvement of the MinD-interacting domain in mediating self-assembly of MinE. Furthermore, residues Ile-72 and Ile-74 on the ␤-strand face of the MinE dimer affected the MinE function through influencing residue Ile-24, suggesting a critical role of the ␤-stranded face of the MinE dimer in regulating the folding propensity of the N-terminal domain. In summary, we provide evidence for self-assembly of MinE on the membrane that may underlie formation of the MinE ring in vivo. The ability of selfassembly may contribute to spatial organization of the Min proteins by providing directionality and coordinating membrane interaction in the system. In addition, biochemical and cell biological characterization of the MinE mutants allows direct examination of the different molecular interactions occurring in the oscillating Min system.
To introduce mutations into minE of pYLS41 [P lac ::minE] (30), pSY1083 [P lac ::minC minD minE] (31), pFX9 [P lac ::gfp-minD minE] (31), and pYLS49 -2 [P lac ::gfp-minC minD minE] (32), long range PCR reactions were performed, and mutations were introduced to the desired sites using primers carrying base substitutions. The mutated DNA fragments were sequenced, excised, and re-ligated into the original plasmids to eliminate additional mutations. Alternatively, the PCR product carrying the desired point mutation was restriction-digested and used to replace the wildtype minE fragment. The same procedures were used to create pSOT181 [P T7 ::minE I24N -his] and pSOT117 [P T7 ::minE I25R -his] from pSOT13 [P T7 ::minE-his]. pSOT190 [P T7 ::minE I74A -his] was created by amplifying the mutant minE fragment from pYLS65 [P lac ::minD minE I72A -gfp] and pYLS66 [P lac ::minD minE I74A -gfp], which was then used to replace the minE fragment in pSOT13 (13) at NcoI and XhoI sites. Mutant plasmids carrying minE-gfp gene fusions were obtained from mutant plasmids derived from pSY1083 and pYLS41 by restriction digestion with BamHI followed by re-ligation. The minC gene was PCR-amplified and cloned into pHTPP15 (29) at the EcoRI and XhoI sites to create pSOT14 [P T7 ::trx-his-minC].
pT25-minD [cyaT25-minD] and pT18-minE [minE-cyaT18] were constructed as described previously (13). For the construction of pT18-minE I24N , pT18-minE I25R and pT18-minE I74A , the mutant minE fragments were created by PCR-amplifying the mutant minE fragment from pSOT181, pSOT117, and pSOT190 and cloned into pT18 at the XhoI and HindIII sites. Fluorescence microscopy, Western blot analysis, and the bacterial two-hybrid assay were performed as described previously (13).
Overproduction and Purification of MinC, MinD, and MinE Fusion Proteins-Procedures for the overproduction and purification of MinE and MinD fusion proteins were described previously (13). The purified MinE carried a His 6 tag at its C terminus and was capable of stimulating MinD ATPase activity and interacting with the membrane (17). In addition, a GFP fusion at the C terminus of MinE localized into a ring-like structure that oscillates from pole-to-pole (3,4), providing evidence that appending a C-terminal tag to MinE retains normal function. To increase solubility of MinE I25R -His, 50 mM L-arginine was added to the elution buffer for affinity purification and to the buffer for gel filtration. The procedure for the purification of the MinC fusion protein was the same as for MinD purification. The purified proteins used in all experiments were clarified by centrifugation at 21,000 ϫ g for 5 min or filtering through 0.22-m syringe filters (Merck KGaA, Darmstadt, Germany) to remove precipitates and were quantified before experiments. The wild-type and mutant MinE 1-31 peptides were synthesized and carried no modification at both ends.
Atomic Force Microscopy-An Agilent 5500 AFM (Agilent Technologies, Inc., Palo Alto, CA) that was equipped with a phase lock loop (PLL) control system (Nanosurf easyPLL plus system; Nanosurf AG, Liestal, Switzerland) to enable frequency modulation detection was used. The system detected laser beam deflection caused by bending of the cantilever. Stiff probes (Nanosensors PPP-RT-NCHR and PPP-NCHAuD; Nanosensors, Neuchatel, Switzerland) were used for detection.
A piece of freshly cleaved mica (6-mm in diameter) was glued onto a metal chip magnetically fixed to the atomic force microscopy (AFM) stage. Preparation of SLBs on mica was based on vesicle fusion as previously described (17), but the fusion reaction was incubated in a sealed chamber and heated at 50°C for 30 min followed by cooling at room temperature for another 30 min. Before applying protein solution, SLBs were examined to ensure the quality. The protein solution prepared in buffer A (20 mM Tris-Cl, pH 7.5, 200 mM sucrose) was added and incubated at room temperature for 2 h. Before installation, 35 l of imaging buffer (20 mM Tris-Cl, pH 7.5, 120 mM KCl, 100 mM sucrose) was applied to immerse the probe in solution. Three to six different positions were imaged for each sample to ensure that the observations were consistent and representative. All scans were performed under constant amplitude and a constant frequency shift at room temperature.
Transmission Electron Microscopy-A 6 M peptide solution was prepared in buffer A. 4 l of peptide solution were spotted onto a 300-mesh grid with Formvar/carbon film (FCF300-Cu, Electron Microscopy Sciences, PA) and incubated at room temperature for 1 min. The EM grid was pretreated in an EMS 100 glow discharge unit (Electron Microscopy Sciences) for 20 s. The sample was stained with 2.5% uranyl acetate for 1.5-2.5 min. For examining liposome tubulation, 12 M MinE (wildtype or mutant) was mixed with an equal volume of 0.4 mg/ml liposomes (400 nm in diameter) in buffer A and incubated for 30 min. 6 l of sample was spotted onto the same type of the EM grid that was glow-discharged for 10 s. The sample was stained with 2.5% uranyl acetate for 40 s. EM images were collected with a Tecnai G2 F20 TWIN electron microscope equipped with a field emission gun (FEI company, Hillsboro, OR) operated at 200 keV and a Gatan UltraScan 4000 CCD Camera (4096 ϫ 4096; 15 m pixel size) and processed with Digital Micrograph software (Version 3.9.5, Gatan).
Congo Red Assay-A 200 M Congo red solution was prepared freshly in 150 mM NaCl and 5 mM potassium phosphate buffer, pH 7.4, and filtered through a 0.22-m filter. 6 M MinE 1-31 was prepared in buffer A and incubated at room temperature for 24 h. The sample was then treated with proteinase K when appropriate. Congo red solution was added to a final concentration of 2.67 M and incubated at room temperature for 30 min before measurements. The UV-visible spectrum was scanned between 200 and 800 nm on a DU 800 spectrophotometer (Beckman, Brea, CA).
Mass Spectrometry-The 1 mg/ml peptide solution was prepared as described above before treating with 1 g/ml proteinase K at room temperature for 16 h. The proteinase K-digested sample was transferred to a Vivaspin 500 column (3,000 MWCO, Sartorius Stedim Biotech GmbH, Göttingen, Germany) and centrifuged at 15,000 ϫ g at room temperature to remove non-fibrillated peptides. The recovered sample was denatured by the addition of an equal volume of 0.01 N HCl and heating at 95°C for 15 min. The sample was applied to another Vivaspin 500 column (MWCO 3000) to recover the peptides disassembled from the denatured fibrils. Subsequently, the peptides in the eluent were analyzed by MALDI-TOF mass spectrometry on a Waters Micromass MALDI micro MX TM system (Waters Corp., Milford, MA). Meanwhile, the peptides in the eluent were separated using an Atlantis dC18 column (4.6 ϫ 100 mm) operated on a HPLC system (Waters Corporation, Milford, MA). The collected peptide fractions were analyzed by MALDI-TOF MS to identify the fraction containing the amyloidogenic peptide. The amyloidogenic peptide in the recovered fraction was further confirmed by protein N-terminal sequencing.
Sedimentation Velocity by Analytical Ultracentrifugation (SV-AUC)-SV-AUC was performed using an XL-A analytical ultracentrifuge equipped with an UV-visible detection system (Beckman Coulter, Fullerton, CA). The flow-through 12-mm centerpiece assembly with quartz windows was loaded with 400-l protein samples and corresponding references before mounting an An-60 Ti rotor for centrifugation at 262,000 ϫ g (at cell center) at 20°C. The radial absorbance at 230 nm was collected at 3-min intervals. A 0.1 mg/ml protein sample was prepared in 20 mM Tris-Cl, pH 7.5, 150 mM KCl, 5 mM DTT, and 200 mM sucrose. The buffer density and viscosity were calculated using the SEDNTERP program (33), and the collected data were analyzed using the SEDFIT software (34).
Cosedimentation Assay-The cosedimentation assay of MinE with liposome was performed as described previously (13,17). The E. coli polar lipids was purchased from Avanti Polar Lipids Ltd. (Alabaster, AL). An extrusion method was used for the preparation of vesicles of uniform size (100 nm) (13). KCl, 5 mM MgCl 2 , and 100 mM sucrose for 15 min at 30°C. The samples were centrifuged at 21,000 ϫ g for 5 min at room temperature and divided into supernatant and pellet fractions.
Electrophoresis was performed following standard procedures described in the Condensed Protocols from Molecular Cloning (35). For quantitation, MinD was separate on a 12% Tris-glycine-SDS gel, and MinE was separated on a 16% Tris-Tricine-SDS gel. After electrophoresis, the gel was stained with 0.1% Coomassie Brilliant Blue R250 prepared in a methanol/ acetic acid/H 2 O (25/10/65) solution for 10 min followed by destaining in water overnight. In general, the Coomassie Blue staining method detects proteins above 0.5 g. We loaded 2 g of MinD (6 M, 10 l) or 0.67 g MinE (6 M, 10 l) in each well for electrophoresis. The gel images were taken using a digital camera and analyzed using the NIH ImageJ software. We took only the band intensity of the intact protein for analysis. The measured intensity of the intact protein from the pellet fraction was divided by the sum intensity of the intact protein measured in both pellet and supernatant to obtain the percentage of pelleted protein.
ATPase Assay-The ATPase activity of MinD was measured using the P i ColorLock TM Gold kit (Innova Biosciences Ltd., Babraham, Cambridge, UK) according to the manufacturer's instructions. Each reaction contained 5 M MinD, 5 M MinE, 0.5 mM ATP, and 0.5 mg/ml liposomes and incubated in a buffer containing 25 mM Tris-Cl, pH 7.5, 50 mM KCl, 5 mM MgCl 2 , and 100 mM sucrose at 30°C for 10 min. The malachite green reagent was added and allowed to stand at room temperature for 30 min before measuring the absorbance at 630 nm.
Circular Dichroism-For CD measurements, reaction mixtures were prepared by mixing MinE 1-31 solution with vesicles to final concentrations of 30 and 100 M. CD spectra were measured in the far UV range (190 -250 nm) on a Jasco J-715 spectrometer (Jasco). The bandwidth and the step resolution were set to 2 and 0.2 nm, respectively. The optical path of the cuvette was 0.1 cm. Three scans were performed for each sample to obtain an averaged spectrum. The collected data were analyzed and processed using the Jascow32 software.

MinE Forms Protein Fibrils on Supported Lipid Bilayers-Al-
though it is known that formation of the MinE ring requires the presence of MinD (3), there was no clear explanation on the ring formation. To examine whether the formation of the MinE ring involves protein self-assembly, we used transmission electron microscopy (TEM) and AFM (36) to examine MinE structures in an in vitro reconstituted system. As control experiments, MinE formed amorphous aggregates under TEM (Fig.  1A) and accumulated irregularly on the mica surface when examined by AFM (Fig. 1B). Although there were vestiges of fibrils in some areas on mica, these appeared to be short range fibrils and were less rigid and might be induced or stabilized by the supporting mica substrate. In addition, SLBs prepared from E. coli polar lipids (phosphatidylethanolamine:phosphatidylglycerol:cardiolipin ϭ 67:23:10%, wt/wt) showed a thickness of 5.0 Ϯ 0.4 nm (n ϭ 31) (Fig. 1, C and D).
We further scanned for MinE structures adsorbed on SLBs that were immersed in solution to preserve MinE activity. Strik-ingly, MinE showed a strong preference for targeting to SLB rims and the junctions between SLB patches at concentrations lower than 0.5 M (Fig. 1, E-G and I-K). The amount of protein accumulation at the membrane edges and the SLB junctions varied from site to site, as seen by the variable intensity differences representing changes in height (Fig. 1, I-K). The high resolution images suggest that short fibrils (16.1 Ϯ 3.5 nm, n ϭ 30) underlay the long range structures at the membrane edges and between the junctions of the membrane patches ( Fig. 1, F, G, J, and K). The short fibrils near the membrane edges were unlikely to have been caused by any tip artifacts for the following reasons. First, the membrane edges were continuous, whereas the fibrils were observed as discontinuous, short sections. Second, the number of parallel short fibrils at the membrane edges varied from site to site. Third, the short fibrils near the membrane edges were consistently observed in all replicate experiments.
The average height of the fibrils bound at the edges was estimated at 1.8 Ϯ 0.6 nm (n ϭ 63) above the supported bilayer. The sharp slope in the line profiles of the edge-associated fibrils indicates no deformation at the bilayer edge (Fig. 1L); thus, the height increase at the edges represents the thickness of the MinE fibrils adsorbed on the bilayer. Another feature of these membrane-associated fibrils is the morphological plasticity, as suggested by their ability to fit into irregular membrane edges and by the high curvature of the long range arrangement of multiple short fibrils. These short fibrillar units likely facilitate modification of the long range organization so as to accommodate shape variations at the membrane edges and junctions.
Self-assembly of the MinD-interacting Domain of MinE-Because the C-terminal domain of MinE is known to form soluble dimers (37), we looked into MinE 1-31 to study the mechanism of fibril formation on the membrane. The TEM results revealed that MinE 1-31 formed both straight and twisted fibrils with different widths (Fig. 2, A and B). Most straight fibrils consisted of two protofibrils, although fibrils of three protofibrils were occasionally seen (Fig. 2A). The width of a protofibril was measured as 5.3 Ϯ 1.3 nm (n ϭ 132) from the straight fibrils. The twisted fibrils showed a periodic distance of 82.0 Ϯ 11.1 nm (n ϭ 108) between twists. The cross-section width of the widest points along the twisted fibrils was measured as 13.0 Ϯ 1.8 nm (n ϭ 43), which may suggest the twisted fibrils were composed of  two tangled protofibrils if they adopted a ribbon-like packing order on a two-dimensional surface or were composed of more than two protofibrils if they were stacked in three dimensions. The observed various fibrillar structures of MinE 1-31 may resemble those at the different maturation states of amyloid fibrils (38), raising the possibility that a segment of the N-terminal domain of MinE may adopt a cross-␤ structure for self-assembly.
We, therefore, investigated whether the anti-parallel ␤-sheet, locating at the dimer interface of the 6␤-stranded conformation (21), represents the subunit for self-assembly (Fig. 2C). One face of this ␤-sheet is enriched with hydrophobic residues Leu-22, Ile-24, and Val-26, which are flanked by glutamic acids (Glu-20 and Glu-28) at both ends. The other face (Gln-23, Ile-25, and Ala-27) shows mixed features, but Arg-21 and Arg-29 at the ends of the ␤-sheet provide positive charges to the surface (Fig. 2E). We hypothesized that self-assembly of MinE 1-31 involves hydrogen bonds between the amyloidogenic subunits into an extended ␤-sheet as well as hydrophobic interactions between two ␤-sheets arranged face-toface, giving rise to a cross-␤ structural arrangement (Fig. 2F).
When Congo red molecules are bound to amyloid fibers, the absorbance maximum generates a spectral shift from 490 to 540 nm. Supporting our hypothesis, the UV spectrum of Congo red mixed with the MinE 1-31 peptide solution gave a signature absorbance at 540 nm, indicating the existence of the cross-␤ structure (Fig. 2G). The spectral shift occurred within a short period of incubation. The spectral shift was also observed for samples after 18 h of incubation and treatment with proteinase K (Fig. 2, H-J), strongly suggesting MinE 1-31 adopts a cross-␤ structure that is resistant to proteinase K digestion. The mass spectrometry analysis on the proteinase K-resistant fragment showed a major fragment of 1069.6 Da, which corresponds to the peptide fragment of 19 KERLQIIVA 27 as confirmed by protein N-terminal sequencing (Fig. 2K). Due to specificity of proteinase K, the residue Lys-19 remains uncertain for its role in fibril formation. The result suggested that residues of this peptide fragment were protected in the cross-␤ spine of the amyloid fibril and thus became resistant to proteinase K digestion.
To further examine our hypothesis of the amyloid fibril formation, we synthesized mutant peptide of MinE 1-31 carrying the I24N or I25R mutation. When the mutant MinE 1-31 peptides were studied by the Congo red assay, a mild shift was observed for the I24N mutant peptide, and the I25R mutant peptide showed a similar spectral shift when compared with the wild-type peptide (Fig. 3A). Under electron microscopy, the I24N mutant peptide failed to form fibrillar structures, but twisted fibrils of the I25R mutant peptide were easily identified (Fig. 3B). The CD spectral analyses were performed to probe for the folding propensity of MinE 1-31 . When dissolved in 20 mM Tris-Cl, pH 7.5, and in the absence of trifluoroethanol (TFE) and liposomes, all MinE 1-31 peptides exhibited spectra indicative of random coils (Fig. 3C). It should be noted that the CD measurement may not report the full folding propensity of a protein, which appeared to be the case for MinE 1-31 . The wildtype peptide and the mutant peptide I24N possessed a helical conformation upon the addition of 30% TFE, with an exception that peptide I25R showed mixed features of both ␣-helical and ␤-stranded conformations (Fig. 3D). In the presence of liposomes, MinE 1-31 showed significant conformational changes, as indicated by CD spectral shifts, but the overall spectrum suggested complicated folds of its secondary structure (17) (Fig.  3E). Interestingly, although the CD spectrum of the mutant peptide I24N did not respond to liposomes, peptide I25R showed a typical ␤-strand spectrum with a distinct minimum at 216 nm. This result indicated that the mutant peptide I25R enhanced the ␤-strand folding propensity in response to 30% TFE as well as liposomes, supporting the observations made from the Congo red assay.
Taken together, the results support that the amyloidogenic core of MinE 1-31 resides in the MinD-interacting domain. The residue Ile-24 supports the hydrophobic interaction between two ␤-sheets; thus, asparagine substitution in residue Ile-24 failed fibril formation. On the contrary, the arginine substitution of residue Ile-25 on the opposite face of the ␤-sheet enhances hydrophilicity to the surface, which promotes hydrophobic interactions between two ␤-sheets of a fibril. The aggregation property of MinE 1-31 may reflect the self-assembly function of the intact MinE under native conditions as discussed below.
MinE Mutants That Affect Cellular Localization of the Min Proteins-In addition to the mutants of I24N and I25R, we report that residues Ile-72 and Ile-74 located on the ␤-stranded face of MinE were involved in regulating functions associated with residue Ile-24 (Fig. 2, C and D). The I72A and I74A mutants of MinE were identified in an alanine scanning experiment aimed at screening for structural determinants of MinE   (14,19).
As functional tests, we found that the MinE mutants I24N, I25R, I72A, I74A, and I74R remained capable of inducing minicell production in the wild-type strain MC1000 (Fig. 4A) but failed to complement the ⌬min strain YLS1 (Fig. 4B). The wildtype MinE expressed from P lac -minE induced a minicelling phenotype with the addition of isopropyl 1-thio-␤-D-galactopyranoside above 10 -25 M. All tested mutants retained their ability to induce minicelling. Production of minicells was observed under both suppressed and inducer-free conditions for the I24N and I25R mutants, indicating aberrant function of the MinE mutants. On the other hand, the wild-type gene effectively restored division at the midcell in YLS1 with the addition of 25 M isopropyl 1-thio-␤-D-galactopyranoside. All mutants failed to complement the minicelling phenotype of YLS1 under all tested conditions. The I25R mutant also showed severe filamentation with the addition of isopropyl 1-thio-␤-D-galactopyranoside, indicating loss of ability to restrict distribution of MinCD to the poles and allowing MinCD to block division over the cell body. The MinE mutants I24N and I74A caused an increasing number of long cells that was associated with increasing concentrations of isopropyl 1-thio-␤-D-galactopyranoside. The observations suggested that all mutant MinE proteins lost the function to mediate the division site placement at midcell.
We further examined cellular localization of MinC, MinD, and MinE by expressing constructs of P lac ::minE-gfp, P lac ::gfp-minC minD minE, P lac ::gfp-minD minE, or P lac ::minC minD minE-gfp in YLS1. In contrast to the dispersed cytosolic localization of the wild-type MinE-GFP in YLS1, mutants I24N, I25R, I72A, I74A, and I74R showed a clear peripheral localization (Fig. 5A), suggesting enhanced membrane binding activities of MinE. Peripheral localization of MinE I24N -GFP and MinE I25R -GFP in a ⌬minCDE background was documented previously (18,30). The fusion protein GFP-MinC normally localizes to the cell pole together with MinD in a MinDE-dependent fashion (7,8). However, the MinE mutants I24N, I72A, I74A, and I74R caused GFP-MinC to mislocalize in the cytosol, and the MinE mutant I25R caused peripheral localization (Fig.  5B). The unexpected cytosolic localization of GFP-MinC in the presence of MinD and MinE mutant I24N, I72A, I74A, or I74R may be caused by the strong membrane binding of MinE over the cell periphery that interferes with MinC binding to the membrane-bound MinD, resulting in the observed cytosolic distribution of MinC. In addition, GFP-MinD localizes to a cell pole when coexpressed with MinE (2), but GFP-MinD showed a uniform peripheral localization pattern when coexpressed with the mutant I24N, I25R, I72A, I74A, or I74R (Fig. 5C). The peripheral localization of MinD and MinE associated with the I25R mutant is consistent with a previous report (30).
MinE-GFP localizes in a ring-like structure capping the medial edge of the MinD polar zone in a MinD-dependent manner (3,4,14). In the presence of the wild-type MinE, 45.1% (n ϭ 153) of the fluorescent cells showed ring-like structures ( Table 2). The fluorescence was also found in one-half of a cell, making a clear contrast between the two ends of a cell (Fig. 5D). The I25R mutant caused a clear peripheral distribution without forming the MinE ring. The MinE mutants I24N, I72A, I74A, and I74R caused a complicated localization pattern, including ring-like structures and peripheral localization and did not show contrast in fluorescence between the two cell halves. The ring structures were found in 22.4% (n ϭ 205) and 33.8% (n ϭ 160) of the fluorescent cells carrying the MinE mutant I24N and I74R, respectively. A similar localization pattern was also identified for the I72A and I74A mutants ( Table 2). The residue Ile-74 appears to associate with more diverse patterns of the MinE-GFP localization than residue Ile-72. The zebra pattern of MinE-GFP fluorescence was identified in 20.6% of the fluorescent cells carrying the I74R mutation. The ring-like structure observed for the mutants could be due to an artifact caused by the GFP fusion that partially masked the mutant phenotype. In addition, 41.9% of cells were non-fluorescent in cells carrying the I74A mutation for an unknown reason, which is not caused by degradation of the mutant protein (Fig. 5E). Taken together, the I24N, I25R, and I74R mutations caused abnormal cellular localization of the three Min proteins.
Biochemical Characterization of the MinE Mutant Proteins-In this section we asked how the I24N, I25R, and I74A mutations affected the biochemical properties of MinE that are required for sustaining the pole-to-pole oscillation of the Min proteins, including membrane binding, CD measurement, liposome deformation, dimerization, MinD-MinE interactions, and the ability to stimulate the MinD ATPase activity. The membrane binding ability of MinE was examined in proteinliposome cosedimentation experiments containing purified protein and liposomes reconstituted from the E. coli polar exact (phosphatidylethanolamine:phosphatidylglycerol:cardiolipin ϭ 68:27:5 mol%). The results showed that the I24N, I25R, and I74A mutants led to a significant increase in MinE binding to the liposomes (Fig. 6, A and B). Here, 35-54% of mutant MinE proteins were observed to pellet with liposomes, compared with just 14% of the wild-type MinE pelleting with liposomes. The enhanced membrane binding ability in the cosedimentation assay was consistent with the distinct peripheral localization of the MinE-GFP mutants (I24N, I25R, and I74A) when expressed alone in the ⌬minCDE background (Fig. 5A).
We demonstrated previously that MinE is capable of inducing liposome deformation and tubulation in an in vitro reconstituted system, which may be due to mechanical force generation accompanying the MinE-membrane interaction (13). We, therefore, inspected the membrane-deformation activity of the mutant MinE proteins by negative-stain TEM (Fig. 6, C-F). Very rare tubulation events were found associating with the I24N mutant (Fig. 6D). Because the MinE mutant I24N is bound to liposome strongly but is defective in self-association based on the observation of the I24N peptide, failure to deform liposomes may be a result of the self-association defect. The results further suggested that membrane binding and self-association  mutants is studied by examining their ability to induce the minicelling phenotype in a wild-type strain MC1000 (A) and  by examining the ability of minCDE to complement the minicelling phenotype of a ⌬min strain YLS1 (B). IPTG, isopropyl 1-thio-␤-D-galactopyranoside.
of MinE can occur independently, but both activities are required to tubulate liposomes. This was in contrast with the I25R and I74A mutants that caused severe deformation of liposomes (Fig. 6, E and F). Among the wild-type and mutant (I25R and I74A) MinE proteins, I74A showed the highest frequency of tubulation. The tubule widths were measured as 53.6 Ϯ 20.6 nm (n ϭ 52), 57.8 Ϯ 10.1 nm (n ϭ 51), and 59.5 Ϯ 16.2 nm (n ϭ 52) for the wild-type, I25R, and I74A mutants, respectively, which showed no statistical differences between each other.
Whether the MinE mutants were defective in dimerization was determined using SV-AUC. SV-AUC analysis showed that all MinE variants were dimeric in solution, whereas a small population of aggregated proteins, which corresponds to an apparent size of trimers or tetramers, was observed for I25R and I74A (Fig. 6G). The results indicated that these two variants were more aggregation-prone in solution. In addition, the three MinE mutants formed amorphous aggregates in solution similar to the wild-type full-length protein as examined by TEM (Fig. 1, A and M-O). We also assayed the ability of MinE to stimulate the MinD-ATPase activity by measuring release of orthophosphate and the ability of MinE to interact with MinD by a bacterial two-hybrid assay (Fig. 6, H and I). There were no major effects of the mutations on the ability of MinE to stimulate the MinD ATPase activity and on the MinD-MinE interactions.
The CD measurements were performed to estimate the secondary structural content of the full-length protein (Fig. 6, J-L). Previously the I24N mutant was proposed to destabilize the dimer interface of the 6␤-stranded conformation (Fig. 2C), facilitating the formation of the 4␤-stranded conformation (Fig. 2D) (18). The CD spectra suggested that the I24N mutant MinE possessed a significantly different folding propensity in buffer with and without TFE (Fig. 6, J and K), which likely reflected a 4␤-stranded conformation due to the mutation. A clear difference between spectra of I74A and other mutants was identified in the presence of TFE (Fig. 6K), suggesting an inherent difference in the folding propensity. The I25R spectra were similar to those of the wild-type protein in buffer with and without TFE (Fig. 6, J and K). In addition, all mutant and wild-type proteins adopted a similar conformation in the presence of liposome (Fig. 6L) even though the I24N mutant lost its ability to induce tubulation.
In summary, the enhanced membrane association of the I24N mutant would suggest the existence of a membrane-associated, 4␤-stranded dimer conformation. In addition, although functional correlations between residues Ile-24 and Ile-72/  Ile-74 were found, the I74A mutant retained its activity to tubulate liposomes. This is explained by the physical involvement of residue Ile-24 in self-assembly of MinE, but Ile-24 remains unchanged in the I74A mutant. Interestingly, the side chains of residues Ile-25, Ile-72, and Ile-74 are aligned in an extended hydrophobic patch on the ␤-stranded face of the 6␤ dimer (Fig.  2C), which may all be involved in sequestration of the N-terminal amphipathic helix. Failure to sequester this amphipathic helix may push MinE for membrane interaction followed by subsequent downstream interactions. In addition, the side chains of Ile-72 and Ile-74 are oriented on the sides of the ␤-stranded face of the 4␤ dimer (Fig. 2D), which may have a role in orienting the MinD-interacting domain through hydropho-bic interactions. Thus residues Ile-72 and Ile-74 likely affect the function of Ile-24 through shifting the interaction of MinE with different partners.

Effect of MinE Mutants on MinD-Membrane Interaction-
We further studied whether and how these MinE mutations affected MinD binding to the membrane in protein-liposome cosedimentation assays (Fig. 7). Under our experimental conditions, 32.6 Ϯ 4.4% of MinD was pelleted with liposomes in the absence of ATP and MinE, indicating MinD binding to the membrane is independent of ATP. The addition of ATP to the system caused a reduction in the amount of pelleted MinD to 22.9 Ϯ 6.0%, which may be caused by the basal hydrolytic activity of MinD in the absence of MinE. Further addition of MinE did not cause a significant change in the amount of the membrane-bound MinD even though increased ATPase activity of MinD was detected upon stimulation by MinE (Fig. 6H). The amount of MinE found in the pellet was reduced to background level in the same reactions (Fig. 7, A and B). The results differed from the previous reports that suggested MinD interacts with the membrane in an ATP-dependent manner, and recruitment of MinE by the membrane-bound MinD leads to release of both MinD and MinE from the membrane (11,39,40). The discrepancy result from the experimental conditions; we used a higher concentration of salt (150 mM KCl) and a crowding agent (12 M BSA) to mimic the cytoplasmic condition in the reactions. The different experimental conditions may have captured differential MinD and MinE interactions at different stages of the oscillation cycle (Fig. 7, E and F) (41). On the other hand, MinC significantly enhanced MinD binding to liposomes by 2.4-fold in the presence of ATP (Fig. 7, C and D). A similar observation was reported by Lackner et al. (11). The addition of MinE to the reaction caused a reduction in the amount of pelleted MinC and MinD, suggesting an important role of MinE in regulating interaction of the MinC⅐MinD-ATP complex with the membrane rather than interaction of MinD with the membrane (Fig. 7, G and H). The MinE mutants I24N, I25R, and I74A were still strongly bound to the membrane in the presence of MinD regardless of the presence or absence of ATP (Fig. 7, A and B). The strong membrane binding ability of the MinE mutants affected the MinD-membrane interaction. In the absence of ATP, the amount of the membrane-bound MinD was reduced to 40.6, 68.6, and 63.1% that of the wild-type level for the I24N, I25R, and I74A mutants, respectively. In the presence of ATP, there was no clear difference in the amount of pelleted MinD for the I24N and I74A mutants, but the amount of pelleted MinD was reduced to 56.1% for the I25R mutant, which implied involvement of the MinE properties of fibril formation and membrane tubulation. The results further support that abnormality of the MinE-membrane interaction can alter the MinD function, including how MinD interacts with the membrane and with MinC. In addition, Ile-25 acts through a different pathway when it is compared with residues Ile-24 and Ile-74 in the MinD-MinE interaction.
Conclusion-The rod-shaped E. coli harbors an oscillating Min system that maintains the spatial distribution of the divi- sion inhibitor MinC at the poles to block aberrant polar division. Although the Min system comprises only three proteins, complex molecular interactions are involved to drive the oscillation cycle. In this work we investigated the selfassembly of MinE on the membrane and characterized MinE mutants that mediate self-assembly and regulate the different interactions of MinE. The results, which are summarized in Table 3, suggest several notions for the molecular bases of the Min system. First, MinE is capable of self-association into fibrils on the membrane, suggesting a self-assembly mechanism in the formation the ring-like structure observed in vivo. Second, the amyloidogenic tendency of the MinDinteracting domain indicates a self-assembly property of MinE on the membrane. Third, the ␤-stranded face of MinE regulates folding propensity of the MinD-interacting domain for different interactions.
A hypothetical model is proposed based on this study. The membrane binding of MinE triggers exposure of the MinDinteracting domain that remains as a ␤-sheet for association with other ␤-sheets from other MinE dimers, causing MinE to oligomerize on the membrane (Fig. 8a). This model of selfassembly would involve formation of a cross-␤ spine that accompanies bending of the oligomer to accommodate the size of the C-terminal domain. Alternatively, the intermediate state preceding self-association may exist that involves refolding of the C-terminal domain to accommodate the structural feature of the cross-␤ spine. The "domain swapping" mechanism (22)(23)(24)(25), which can lead to oligomerization of a protein, is also possible for the formation of the amyloid aggregates of MinE. In this model the MinD-interacting domains would be exchanged between adjacent MinE dimers to form hydrogen bonds between interdimer ␤ strands (Fig. 8b). MinD at the medial edge of the polar zone likely mediates the long range organization of the MinE oligomers, explaining the dependence of MinD on the formation of the MinE ring. The pathway in this model is equally important in a recent model addressing the MinD-MinE interaction, which suggests MinE exists in a 6␤-stranded dimer conformation in solution with the N-terminal membrane-targeting sequence folded into an amphipathic helix and is sequestered via a hydrophobic interaction with residue Ile-25 on the ␤-stranded face (18,21). In this model the interaction between the cytosolic 6␤-stranded MinE dimer and MinD triggers exposure of the MinD-interacting domain from the dimer interface followed  The ϩ symbol indicates that activity was detected. The wild-type activity is normalized as 100%; the mutant activity is relative to that of the wild-type. The Ϫ symbol indicates that no activity was detected. b Minicelling phenotype. c Filamentation phenotype. d The ϩ symbol indicates that protein was detected in the pellet fraction. The abundance of the wild-type protein found in the pellet fraction is normalized as 100%; the abundance of the mutant protein is relative to that of the wild-type.
by refolding of the MinD-interacting domain into an ␣-helix for interaction with MinD (18). An interesting question arising from this work is the kinetics property between assembly and disassembly of the ring-like structure in the form of amyloid aggregates during oscillation. Further investigation is required to address this question.