Central Domain of DivIB Caps the C-terminal Regions of the FtsL/DivIC Coiled-coil Rod*

DivIB(FtsQ), FtsL, and DivIC(FtsB) are enigmatic membrane proteins that are central to the process of bacterial cell division. DivIB(FtsQ) is dispensable in specific conditions in some species, and appears to be absent in other bacterial species. The presence of FtsL and DivIC(FtsB) appears to be conserved despite very low sequence conservation. The three proteins form a complex at the division site, FtsL and DivIC(FtsB) being associated through their extracellular coiled-coil region. We report here structural investigations by NMR, small-angle neutron and x-ray scattering, and interaction studies by surface plasmon resonance, of the complex of DivIB, FtsL, and DivIC from Streptococcus pneumoniae, using soluble truncated forms of the proteins. We found that one side of the “bean”-shaped central β-domain of DivIB interacts with the C-terminal regions of the dimer of FtsL and DivIC. This finding is corroborated by sequence comparisons across bacterial genomes. Indeed, DivIB is absent from species with shorter FtsL and DivIC proteins that have an extracellular domain consisting only of the coiled-coil segment without C-terminal conserved regions (Campylobacterales). We propose that the main role of the interaction of DivIB with FtsL and DivIC is to help the formation, or to stabilize, the coiled-coil of the latter proteins. The coiled-coil of FtsL and DivIC, itself or with transmembrane regions, could be free to interact with other partners.

Cell division is one of the defining features of life. Understanding the division of bacteria is also required to find novel antibiotic strategies. Numerous studies, carried out mostly with the model organisms Escherichia coli and Bacillus subtilis have uncovered several components of the divisome, which can be defined as the ensemble of proteins localized at the division site and participating in the process. Comparison of genomes and deletion studies indicate that the core of the divisome comprises eight conserved, mostly essential proteins: FtsZ, FtsA, FtsK, FtsQ(DivIB), FtsL, FtsB(DivIC), FtsW, and FtsI. Fts nomenclature applies to Gram-negative organisms, whereas Div nomenclature applies to Gram-positive bacteria. These proteins are listed here in the conditional order of their recruitment to the division site of E. coli (1)(2)(3)(4).
Processes in which they participate have been attributed to several division proteins. FtsZ forms polymers with an annular distribution on the cytoplasmic side of the membrane and governs the recruitment of the other proteins. FtsA may mediate the interaction of FtsZ with the membrane. FtsK participates to the resolution of chromosome dimers, and possibly to the membrane fission. FtsI, and likely FtsW, participate to septal cell wall formation (1)(2)(3)(4). In contrast, the roles of FtsQ(DivIB), FtsL, and FtsB(DivIC) have not been firmly linked to any particular process.
FtsQ(DivIB), FtsL, and FtsB(DivIC) are positioned in the middle of the conditional order of recruitment in E. coli and B. subtilis. When the temporality of the recruitment was examined, FtsQ(DivIB) was found to belong to the late recruits, together with the proteins involved in cell wall assembly (5). In E. coli, the presence of FtsL and FtsB at the division site is mutually dependent, and their localization depends on that of FtsQ (6,7). In B. subtilis, the presence of FtsL and DivIC at mid-cell depends on that of DivIB, at the temperature at which DivIB is essential, and reciprocally (8,9). A complex comprising FtsQ, FtsL, and FtsB was isolated from E. coli by co-immunoprecipitation (10), and reconstituted in vitro with recombinant soluble forms of pneumococcal DivIB, FtsL, and DivIC (11). The interaction of the three proteins was also confirmed by yeast and bacterial triple hybrid (12,13).
The genes ftsL and ftsB(divIC) are essential in E. coli and B. subtilis (6, 14 -16) and presumably in Streptococcus pneumoniae (17). The essentiality of ftsQ(divIB) in laboratory conditions varies between species. The gene ftsQ is essential in E. coli (18), but divIB is essential only at high temperatures in B. subtilis (9,19), or in a chemically defined medium in S. pneumoniae (17). Under these conditions, the essentiality of DivIB appears to be a consequence of the protection from proteolysis that it provides to FtsL (8,17).
FtsQ(DivIB), FtsL, and FtsB(DivIC) are bitopic membrane proteins with an N-terminal cytoplasmic region, a single transmembrane segment, and an extracytoplasmic region. The extracellular part is necessary and sufficient for the localization and function of FtsQ(DivIB), provided that it is anchored to the membrane (e.g. Refs. 20 and 21)), although the transmembrane segment also contributes to the localization (22,23). The extracellular part is organized in three regions termed ␣, ␤, and ␥. The crystal structure of a region consisting of the ␣and ␤-domains was solved for FtsQ from E. coli and Yersinia enterocolitica (24). The ␣-domain, comprising about 70 amino acids proximal to the cytoplasmic membrane, corresponds to the POTRA (for polypeptide transport-associated) domain first identified by sequence analysis and proposed to function as a molecular chaperone (25). The ␣and ␤-domains form the conserved region of the FtsQ(DivIB) protein. The ␥-region constitutes a C-terminal tail. It is highly variable in length and sequence and predicted to be unfolded. The ␥-region was not observed in the structures from E. coli and Y. entercolitica, thus confirming its flexible nature (24).
The ␣-domain in the recombinant soluble form of the extracellular part of DivIB from Geobacillus stearothermophilus was digested by trypsin and therefore considered to be largely unfolded (26). The ␥-region was also removed by trypsin digestion, together with a C-terminal fragment of the ␤-domain. The structure of the resulting shorter ␤-domain from G. stearothermophilus was solved by NMR (26) and lacks the two C-terminal ␤-strands.
Localization epitopes have been identified in the transmembrane segment, the ␣-domain, and a region encompassing the C-terminal part of the ␤-domain and ␥-tail of DivIB from B. subtilis (23). Likewise in E. coli, a region in the ␣-domain is required for localization of FtsQ, whereas the C-terminal region of the ␤-domain and the last ␣-helix are required for recruitment of FtsL and FtsB (24). In S. pneumoniae, the essentiality of DivIB in defined medium was found to reside in the C-terminal region of the ␤-domain (17).
No experimental structure is known for FtsL or FtsB(DivIC). Both are small proteins comprising between 90 and 140 amino acids. The number of residues is sometimes larger, as in Mycobacterium tuberculosis (384 for FtsL and 228 for FtsB), due to N-and/or C-terminal extensions consisting of mostly charged and polar amino acids or proline-rich sequences. The major part of FtsL or FtsB(DivIC) is extracellular and contains a region proximal to the transmembrane segment, predicted to form a coiled-coil of about five heptads. Coiled-coil helices associate longitudinally to mediate protein association. It is possible that the coiled-coil helices are continuations of the transmembrane helices, although a proline (known to break helices) is present in some species between the two segments. Following the coiledcoil region is a 25-35-residue long C-terminal region in both FtsL and DivIC(FtsB). This region was recently shown in FtsB to be required for interaction with FtsQ in E. coli (27).
We report here the results of structural studies in solution of a ternary complex consisting of the ␤and ␥-segments of DivIB, and a constrained dimer of the extracellular parts of FtsL and DivIC from S. pneumoniae. Despite the coiled-coil predictions, the recombinant extracellular domains of FtsL and DivIC did not interact in vitro (11,28). Forced dimerization was obtained by fusion with artificial coiled-coil peptides k5 and e5 (35 residues long), which are known to form a heterodimer due to their complementarity of charge, with a nanomolar dissociation constant (29). The k5-and e5-coils were fused to the extracellular domain of FtsL and DivIC, to give rise to KL and EC fusion proteins, respectively. The constrained dimer (KL/EC) was shown to interact with the extracellular part of DivIB (DivIB ext ), yielding a soluble complex amenable to structural studies (11).
The overall shape of the complex and its constituents was probed using small-angle x-ray scattering (SAXS) 2 and smallangle neutron scattering (SANS). NMR was used to investigate the interface between the proteins by chemical shift mapping. The interaction was further investigated using surface plasmon resonance with truncated forms of the proteins. The complex of DivIB, FtsL, and DivIC is formed by the interaction of one face of the ␤-domain of DivIB with the C-terminal regions of FtsL and DivIC, at the tip of an elongated rod formed by the coiled-coil segments. The ␣-domain of DivIB and the coiledcoil regions of FtsL and DivIC remain free to interact with other proteins of the division apparatus.

EXPERIMENTAL PROCEDURES
Plasmid Constructions-Proteins were produced in E. coli from expression plasmids inducible with isopropyl ␤-D-1-thio-galactopyranoside. The sequence encoding DivIB ext preceded by the sequence for a tobacco-etched virus protease cleavage site (TEV) was introduced into pGEX-4T1 (GE Healthcare) to produce the fusion protein glutathione S-transferase (GST)-TEV-DivIB ext (11). The sequence encoding the ␤␥-fragment (residues 224 to 396) was similarly introduced into pGEX-4T1 to produce GST-TEV-␤␥. The sequence encoding the ␤-domain (residues 224 to 361) following a TEV site was introduced into pLX06 (ProteinЈeXpert) to produce the fusion protein GST-TEV-␤.
To produce the KL protein (residues 44 to 105 of FtsL fused following the k5 peptide) with an N-terminal poly-His tag, and a tryptophan to generate absorption at 280 nm, the codon TGG was introduced by site-directed mutagenesis into the pETduetbased (Novagen) plasmid described previously (11). To produce KL* (as KL but ending at residue 82 of FtsL), the stop codon TAG was introduced by site-directed mutagenesis. To produce EC (residues 55 to 122 of DivIC fused following the e5 peptide) with an N-terminal Strep tag (IBA), the sequence encoding Strep-EC was amplified from the plasmid described previously encoding a poly-His-tagged EC (11) and introduced in pET30 (Novagen). The sequence encoding a poly-His tag followed by a TEV site and EC was similarly introduced into pET30 (Novagen), whereas the sequence encoding a poly-His tag followed by a TEV site and EC* (as EC but ending at residue 93) was introduced into pLIM09 (RoBioMol). A plasmid was also constructed for coexpression of KL and EC by subcloning the sequence encoding the Strep-tagged EC into the pETduet-based plasmid encoding the poly-His-tagged Trp-containing KL.
Protein Production and Purification-Unlabeled proteins were expressed in E. coli BL21(DE3) RIL in flasks or in a fermentor (Minifors, Infors) in Luria broth or Terrific broth and the appropriate antibiotic at 37°C following the addition of 0.5 mM isopropyl ␤-D-1-thio-galactopyranoside. For the produc-tion of 15 N-labeled DivIB ext , cells were grown in a fermentor at 28°C in minimal medium (KH 2 PO 4 , 3 g/liter; Na 2 HPO 4 , 6 g/liter; NaCl, 0.5 g/liter; DivIB ext , ␤, and ␤␥ were purified in a first step by glutathione affinity chromatography. Cell lysates in 20 mM Tris, pH 8, 500 mM NaCl, containing the GST-TEV-DivIB ext , GST-TEV-␤, or GST-TEV-␤␥ fusion proteins were loaded onto glutathione-Sepharose resin (GE Healthcare). The fusion proteins were cleaved on the resin by incubation at 37°C with the poly-Histagged TEV protease and recovered by washing with 20 mM Tris, pH 8, 150 mM NaCl. The TEV protease was then removed by binding on Ni-NTA resin (Qiagen). This procedure was sufficient for the 15  Further purification steps were also applied before scattering experiments. DivIB ext , ␤, and ␤␥ proteins from the affinity purification were dialyzed against 20 mM Tris, pH 8, to lower the NaCl concentration to 15 mM, prior to loading onto a Resource Q resin (GE Healthcare) and elution with a NaCl gradient from 15 to 500 mM in 10 column volumes. Fractions of interest were pooled and concentrated with Amicon Ultra devices (Millipore) prior to size exclusion chromatography on a Superdex 75 column (10 ϫ 300 mm) (GE Healthcare). Homogenous fractions (as judged by SDS-and nondenaturing polyacrylamide gel electrophoresis) were pooled, dialyzed against 20 mM Tris, pH 8, 100 mM NaCl, and finally concentrated with Amicon Ultra devices.
The protein EC with a Strep-tag was purified from the cell lysate in 20 mM Tris, pH 8, 150 mM NaCl by affinity chromatography on a streptactin resin (IBA). The protein was eluted with 2.5 mM desthiobiotine in the same buffer.
The proteins EC, KL, EC*, and KL* with poly-His tags were purified from cell lysates in 20 mM Tris, pH 8, 500 mM NaCl, 20 mM imidazole, by affinity chromatography on Ni-NTA resin (Qiagen). Proteins were eluted with an imidazole gradient of 20 to 500 mM in 10 column volumes.
The (KL/EC), (KL*/EC), and (KL/EC*) complexes, with both partners harboring poly-His tags, were prepared from proteins purified as described above. The two proteins were mixed in equivalent amounts. Urea was added to reach a concentration of 8 M. The urea was then removed by dialysis three times against 40 volumes of 20 mM Tris, pH 8, 150 mM NaCl.
The (KL/EC) complex, where EC has a Strep tag and KL a poly-His tag, was prepared from a lysate of cells co-expressing both proteins. The complex was isolated by two successive steps of affinity chromatography in 20 mM Tris, pH 8, 150 mM NaCl, first on Ni-NTA resin eluted with an imidazole gradient from 20 to 500 mM, then on streptactin eluted with 2.5 mM desthiobiotine. Following concentration with an Amicon Ultra device, the complex was further purified by size exclusion chromatography on a Superdex S200 column (16 ϫ 600 mm) equilibrated with 20 mM Tris, pH 8, 150 mM NaCl. For the transverse relation optimized spectroscopy NMR experiment, the buffer was exchanged by desalting on a PD10 column (GE Healthcare) against 100 mM ammonium acetate, pH 7, prior to lyophilization.
Determination of Protein Concentrations-Protein concentrations were determined by the absorbance at 280 nm using the following theoretical extinction coefficient: 5500 M Ϫ1 cm Ϫ1 for poly-His-tagged KL and KL*, 17 M. Spectra were recorded on a 800 MHz Varian spectrometer equipped with a cold probe. For backbone assignment, a sample of the uniformly 13 C/ 15 N-labeled ␤-domain at 0.8 M was prepared and data were collected at 27°C in a 20 mM sodium phosphate buffer (90% (v/v) H 2 O, 10% (v/v) D 2 O) at pH 6 with 150 mM NaCl. A set of high field triple resonance experiments (HNCO, HNC␣CO, HNC␣C␤, and C␤C␣CONH) where acquired and assignment was done using the CO, C ␣ , and C ␤ resonances. First assignment was obtained by an automatic approach using MARS (30) and was then refined manually.
The binding of (KL/EC) to the ␤-domain was studied by transverse relation optimized spectroscopy-type 1 H-15 N correlation experiments. NMR data were collected at 25°C with 32 transients and 1024 ( 1 H) ϫ 64 ( 15 N) complex points. A reference spectrum was taken on a 0.4 mM uniformly 13 C/ 15 N-labeled deuterated ␤-domain sample and lyophilized (KL/EC) was added to a concentration of 0.5 mM.
SAXS and SANS Experiments and Data Processing-The scattered intensities I(Q) were analyzed as a function of the wave vector transfer Q, with Equation 1, where 2 is the scattering angle and the x-ray or neutron wavelength.
The SAXS measurements were performed on the ID02 beamline (ID2 high brilliance beamline) at the European Syn- OCTOBER 2, 2009 • VOLUME 284 • NUMBER 40 chrotron Radiation Facility (Grenoble, France) at a sample to detector distance of 2 m to cover the Q-range from 0.008 to 0.28 Å Ϫ1 . The x-ray wavelength was 1 Å and data were collected by a high sensitivity CCD (FReLoN) detector placed in an evacuated flight tube. Solutions were loaded in a flow-through quartz capillary cell (diameter, 2 mm; wall thickness, 10 m) at 15°C. The radiation damage was checked with 10 successive exposure times of 0.1 s each on a single part of the sample. Longer exposure time led to damage. Therefore, we acquired and averaged 10 curves of 0.1-s exposure time on different parts of the sample that was pushed through the capillary cell. The two-dimensional diffraction patterns were normalized to an absolute scale and azimuthally averaged to obtain the intensity profiles I(Q), and the solvent background was subtracted using Fit2D.

Structure of the DivIB-FtsL-DivIC Complex
SANS was carried out on the large dynamic range smallangle diffractometer D22 at the Institut Laue-Langevin (Grenoble, France). 200-l Samples in H 2 O buffers were measured in Hellma QS 1-mm quartz cells at 4°C. (Data sets from samples in D 2 O could not be used due to aggregation problems.) The sample to detector distances were 2 and 8 m with an incident neutron wavelength of 6 Å to cover a Q-range from 0.025 to 0.30 Å Ϫ1 . The sample and buffer raw two-dimensional intensities (after masking inappropriate data points) were normalized to the incoming neutron flux and corrected for detector efficiency (H 2 O reference), electronic background (B 4 C standard), and sample holder scattering (empty quartz cuvette), using RNILS (34). Corrected two-dimensional sample and buffer intensities were summed up azimuthally into one-dimensional I(Q) intensities using SPOLLY (34). Buffer intensities were then subtracted using PRIMUS (35).
Determination of Molecular Weights-Molecular masses were determined by SAXS relatively to hen egg white lysozyme using Equation 2, where I(0) is the forward scattering and C is the protein concentration in g/liter. The factor ƒ 2 ϭ (83 Ϯ 1)10 Ϫ6 mol L g Ϫ2 was determined using hen egg white lysozyme at a concentration of 10.0 g/liter (36).
Molecular weights were determined on an absolute scale by SANS using Equation 3 (37), where I(0) is the coherent macromolecular forward scattering (extracted by the Guinier approximation, see Equation 4), I inc (0) is the incoherent forward scattering from H 2 O, T s and T are the transmissions of the sample and H 2 O, respectively. C is the protein concentration in g/liter, t is the sample thickness in cm, ƒ is a correction factor for the anisotropicity of the solvent scattering as a function of neutron wavelength. b i are the scattering lengths of the protein atoms in cm, s is the solvent scattering density in cm Ϫ2 , V is the protein excluded volume, and N A is the Avogadro constant.
Determination of the Radii of Gyration-We extracted the radii of gyration R g of all samples by a Guinier analysis using Equation 4 (38), where I(0) is the intensity scattered in the forward direction.
The I(0) and R g values were extracted by Guinier approximation using the program PRIMUS (35). The validity of the approximation was checked a posteriori in each case and was always fulfilled (R g Q (max) Ͻ 1). The cross-sectional radius of gyration R C was determined according to Equation 5, with I C (Q) ϭ I(Q)Q/L, with a graphical fit using the program PRIMUS (35). The good agreement between the radii of gyration and the maximum extension of the particle (L ϭ D max ) was checked using the relationship in Equation 6.
Distance Distribution Functions and ab Initio Modeling-The distance distribution functions P(r) were determined by the program GNOM (39) from SAXS and SANS curves. The boundary conditions P(0) ϭ P(D max ) ϭ 0 were imposed in each case, D max being the maximum extension of the particles. Low-resolution envelopes were calculated from the P(r) functions with DAM-MIN (40).
Rigid Body Modeling-The experimental SANS data from the complex ␤␥(KL/EC) were modeled using the ␤␥ ab initio model and the coiled-coil structure of cortexilin I (Protein Data Bank entry 1D7M) as rigid bodies using the program MASSHA with default parameters (41). In addition, several arrangements of both rigid bodies were generated manually using MOLMOL (42) and scored against the SANS curve using CRYSON (43).
Calculation of Scattering Intensities from Known Structures-Theoretical scattering curves I(Q) of the E. coli and G. stearothermophilus ␤-domains were calculated from PDB entries 2VH1 and 1YR1 using the program CRYSOL (44) with default parameters.
Surface Plasmon Resonance Measurements-Measurements were performed with a Biacore TM 3000 instrument. All samples were dialyzed against 10 mM Hepes, pH 7.5, 150 mM NaCl, 50 mM EDTA, and 0.005% P20. The dialysis buffer was used as running solution throughout. KL/EC, KL*/EC, and KL/EC* mixtures (with all proteins carrying an N-terminal poly-His tag) were attached to a NTA sensor chip activated with 10 l of 500 M NiCl 2 . Experiments were at room temperature with a 20-l/ min flow rate. The (KL/EC), (KL*/EC), and (KL/EC*) dimers (ligands) were injected at the concentration of 100 nM. The DivIB ext , ␤, and ␤␥ proteins (analytes) were injected at concentrations ranging from 5 M to 1.5 mM. For each experiment, interaction curves recorded for the running buffer on the ligand, and for the analyte on the naked sensor chip, were subtracted. Sensorgrams were analyzed with the program BIAevaluation 3.0.

RESULTS
Sequence Comparisons-The coiled-coil regions of FtsL and DivIC(FtsB) are usually followed by 25 to 30 amino acids that are C-terminal, except in species with long additional unfolded extensions. There is little sequence conservation between species, and it is sometimes difficult to identify the genes encoding FtsL or DivIC(FtsB) using sequence comparison methods. Instead, it is best to look for genes encoding proteins with a single transmembrane segment followed by a predicted coiledcoil region, in the proper chromosomal environment. The ftsL gene is usually following mraW, whereas divIC(ftsB) is usually close to that of enolase.
A set of FtsL and DivIC(FtsB) sequences were aligned using the program CLUSTALW. The transmembrane region was predicted using whole alignments with the program TMAP. The coiled-coil regions were predicted using PAIRCOIL2. The alignments are presented in supplemental Figs. S1 and S2.
Despite the absence of overall sequence conservation of FtsL, three hydrophobic amino acids (a Val/Leu/Ile, an Ala, and a Leu/Met) were conserved in the region following the predicted coiled-coil (supplemental Fig. S1). In FtsB(DivIC), a basic (Arg) and an acidic (Glu) residue were conserved in the region following the coiled-coil (supplemental Fig. S2).
It is noteworthy that the extracellular region of both FtsL and FtsB are shorter in Campylobacterales, with very little sequence following the predicted coiled-coil region, and lack the conserved residues mentioned above. Remarkably, Campylobacterales also appear devoid of FtsQ, suggesting that the C-terminal regions of FtsL and FtsB(DivIC) containing the conserved residues might mediate the interation with (FtsQ)DivIB.
Limited Proteolysis-To probe the domain organization of DivIB from S. pneumoniae, we submitted DivIB ext (corresponding to residues 150 to 396) to trypsin digestion. A time course monitored by Coomassie-stained SDS-polyacrylamide gel electrophoresis is shown in Fig. 1. DivIB ext (28,177 Da) migrated as a 28-kDa protein. After 50 min of digestion with trypsin, a major stable fragment migrated as a 15-kDa protein. The mass was determined by electrospray mass spectrometry to be 15,842 Ϯ 1 Da. N-terminal sequencing yielded the sequence VKEYDIVA. Combining the mass and sequence data indicated that the fragment spans residues 220 to 361 of pneumococcal DivIB (15,843 Da). This fragment overlaps the ␤-domain defined by the structure of FtsQ from E. coli, and is somewhat longer on the C-terminal side than the ␤-domain defined by partial proteolysis of DivIB from G. stearothermophilus. The trypsin-sensitive regions 150 to 219 and 362 to 396 correspond to regions ␣ and ␥, respectively.
For further studies, two truncated forms of pneumococcal DivIB that we termed ␤ and ␤␥, corresponding to residues 224 to 361 and 224 to 396 C terminus, respectively, were purified. We decided to produce proteins with a truncated N terminus with respect to our proteolysis results to exclude the fully conserved Glu-222, which is absent from the ␤-domain from G. stearothermophilus, and appears to be part of the linker between the ␣and ␤-domains from E. coli and Y. enterocolitica. Both ␤ and ␤␥ contain an additional N-terminal glycine resulting from cleavage of the fusion with the GST using the TEV protease. The [ 15 N-1 H]HSQC spectrum of the ␤-domain showed about 100 well dispersed peaks superimposable to those of DivIB ext (not shown), without an accumulation of resonances in the 120-or 8-ppm regions. The comparison of the spectra of DivIB ext and ␤ showed that the trypsin-sensitive ␣and ␥-segments were largely unstructured.
When an excess of unlabeled (KL/EC) was added to 15 N-DivIB ext , at a concentration where most of the DivIB ext should be interacting with the dimer, the [ 15 N-1 H]HSQC spectrum showed the disappearance of nearly all the dispersed peaks and the persistence of the peaks due to unfolded parts of the protein (not shown). We interpreted these observations as indicating that the (KL/EC) dimer interacts with the ␤-domain of DivIB, without important contact with the ␣or ␥-regions. When the same experiment was performed with only the 15 N-␤-domain, all the peaks disappeared. The disappearance of the peaks from the ␤-domain upon association with (KL/EC) could be explained by a greater transverse relaxation resulting from a change in the hydrodynamic behavior of the protein. The ␤(KL/ EC) complex likely assumes a non-compact shape that drastically reduces its rotational diffusion. If the complex were compact, the 2-fold difference in size between the free ␤-domain and ␤(KL/EC) would not reduce the rotational diffusion to the point of extinguishing the signal. The persistence of the signal of the ␣and ␥-regions observed on the complete 15 N-DivIB ext in the presence of (KL/EC) indicates that these do not participate in the interaction, and that they retain their flexibility.
The disappearance of the signal upon formation of the ternary complex precluded further use of the 15 N-labeled samples. To minimize the translational relaxation, and thus compensate the loss of signal due to the loss of rotational diffusion upon formation of the ternary complex, we prepared a 15  H]HSQC spectrum, 97 could be assigned, including 27 that experienced modification of their chemical shift upon binding of (KL/EC). The attribution is nearly complete for the first three quarters of the ␤-domain (residues 224 to 323) but could not be achieved for residues 324 to 352 (Fig. 2B), probably due to the dynamics of conformational exchanges of the region. Secondary structure propensities obtained from the 13 C ␣ and 13 C ␤ chemical shifts of the pneumococcal protein, compared with the secondary structures of the E. coli and G. stearothermophilus proteins, are given in supplemental Fig. S3.
Small-angle X-ray Scattering Investigation of the Extracellular Part of DivIB and (KL/EC)-To obtain low-resolution structural information on the extracellular part of DivIB in solution, a series of small-angle x-ray scattering measurements were performed on various truncated forms of the protein.
Small-angle scattering curves provide information on the mass and shape of the scattering particle. The mass is proportional to the intensity scattered in the forward direction (I(0)) provided that the sample is dilute and that there is a single scattering species, and can be extracted from SAXS and SANS data using Equations 2 and 3, respectively (see "Experimental Procedures"). The shape of the scattering curve I(Q) is related to the shape of the particle and yields the radius of gyration, R g , at the smallest scattering angles (Guinier approximation, see Equation 4 under "Experimental Procedures"). The distribution of distances between scattering points of the particle (termed distance distribution function P(r)) can be extracted from the whole scattering curve. Combined, the radius of gyration, the profile of the distance distribution function, and the molecular mass hint at the shape of the particle (e.g. elongated versus compact). Theoretical distance distribution functions and scattering curves can be unambiguously calculated for any particle of known atomic structure. The reverse is not true, but three-dimensional models of a particle can be constructed by iterative numerical simulation.
The complete extracellular part of DivIB ext produced scattering curves with very high forward scattering I(0) (not shown), indicating that samples were not monodisperse but contained oligomers or aggregates. Masses determined from the I(0) values increased with the protein concentration. The presence of aggregates of DivIB ext were also observed by analytical ultracentrifugation (not shown). The formation of aggregates was not unexpected given that DivIB ext is partly unfolded, but precluded the use of SAXS or SANS for further investigation.
The shape of the SAXS curves obtained for the ␤ and ␤␥ proteins showed no variation with the concentration (between 117 and 685 M for ␤, between 33 and 313 M for ␤␥). The apparent masses determined from the I(0) values were consistent with the theoretical masses. The measured mass for ␤ was 16 Ϯ 2 kDa for an expected mass, M theor ϭ 15 kDa, that of ␤␥ was 25 Ϯ 3 kDa for an expected mass, M theor ϭ 20 kDa. The shape of the scattering curves and the good agreement between measured and calculated masses showed that the samples were largely monodisperse, as observed by analytical ultracentrifugation (not shown).
For comparison, theoretical scattering curves were calculated for the ␤-domain of DivIB from G. stearothermophilus (residues 115 to 233) and FtsQ from E. coli (residues 126 to 260) using the program CRYSOL (44). The normalized experimental and theoretical curves are shown in Fig. 3A. The experimental SAXS curve of the ␤-domain from S. pneumoniae is very similar to that calculated for the E. coli orthologue, indicating that both protein domains have the same overall shape. However, the radii of gyration R g of both ␤-domains are somewhat different (20.8 Ϯ 1 and 17.4 Å, for S. pneumoniae and E. coli, respectively), despite having nearly the same mass (15, 381 and 15,317 Da). The pneumococcal ␤-domain is therefore likely more elongated, as shown by the distance distribution function P(r) in Fig. 3B, which displays more long distances.
In contrast, the calculated scattering curve from the ␤-domain of G. stearothermophilus deviates significantly from those from S. pneumoniae and E. coli (Fig. 3A). At small Q-values, it decreases less rapidly, whereas the intensity decreases strongly at larger Q-values (Q Ͼ 0.15 Å Ϫ1 ), suggesting that the ␤-domain from G. stearothermophilus is more compact than its pneumococcal orthologue, with a radius of gyration R g ϭ 15.3 Å.
The experimental SAXS curve of the ␤␥-fragment decreases more rapidly at small Q-values than those of the ␤-domains from S. pneumoniae and E. coli, indicating a more elongated conformation (Fig. 3A). At large Q-values (Q Ͼ 0.13 Å Ϫ1 ) curves of ␤and ␤␥-proteins look similar. The P(r) of the ␤␥-fragment is nearly identical to that of the ␤-domain, with one shoulder at greater distances. This comparison suggests that the ␤␥-protein has the same shape as the ␤-domain with a longitudinal extension.
Ab initio models of the ␤and ␤␥-proteins were calculated from the distance distribution functions. Five independent models were produced and were consistent (Figs. 3C and supplemental S4). The resolution of the ␤ models is coarser than that of ␤␥, as the data were noisier at large Q-values. One representative model of each pneumococcal truncated form (␤ and ␤␥) is shown in Fig. 3C, alongside structures of the ␤-domain from G. stearothermophilus and E. coli. The ␤-domain of pneumococcal DivIB appears to be elongated and slightly curved, or "bean"-shaped. Models of the ␤␥ protein fragment comprise a bean-shaped domain and an additional smaller globular domain attached by a short stalk at one extremity of the bean, which likely constitutes the ␥-tail. The ␥-region is unstructured as judged by its sensitivity to trypsin and the NMR data. The smaller globular domain of the ␤␥ model could represent conformational sampling by the unfolded ␥-tail.
As expected from direct comparison of the SAXS and theoretical scattering curves, the pneumococcal ␤-domain has an overall shape similar to that of the E. coli ␤-domain. The crystal The noise in the data at higher Q-values is due to the short exposure time used to limit radiation damage. B, distance distribution functions P(r) corresponding to the scattering curves shown in A. C, envelope of the high-resolution structures of the ␤-domains from G. stearothermophilus in gray and E. coli in green (from the NMR structure PDB code 1YR1, and x-ray structure PDB code 2VH1, respectively), and low-resolution models calculated ab initio from the distance distribution functions of the pneumococcal ␤-domain (in blue) and ␤␥-fragment (in red). D, superimposition of the envelope of the low-resolution SAXS model of the ␤␥-fragment from S. pneumoniae and the ribbon representation of the ␤-domain from E. coli. OCTOBER 2, 2009 • VOLUME 284 • NUMBER 40 structure of the E. coli ␤-domain can be placed easily within the model envelope of the bean-shaped domain of the ␤␥-fragment (Fig. 3D).

Structure of the DivIB-FtsL-DivIC Complex
Relative molecular mass measurements by SAXS of the (KL/ EC) artificially constrained dimer at various concentrations indicated that the dimer was self-associating to form a tetramer (KL/EC) 2 with a dissociation constant comprised between 5 and 50 M (not shown). We did not pursue structural characterization of the (KL/EC) 2 species as it is likely an artifact resulting from higher order association of the k5/e5 coiled-coil, which is known to form (k5/e5) 2 tetramers at high concentrations (29). Note that hydrodynamic techniques such as size exclusion chromatography, dynamic light scattering, or analytical ultracentrifugation failed to reveal the formation of the tetramer (KL/EC) 2 , presumably because the (KL/EC) and (KL/ EC) 2 particles have comparable hydrodynamical properties despite having a 2-fold mass difference, due to the elongated, highly anisotropic nature of the coiled-coils.
Small-angle Neutron and X-ray Scattering Investigations of the Ternary Complex ␤␥/KL/EC-At high concentrations of ␤␥/KL/EC (Ͼ10 M), necessary to generate high quality scattering data, we could have expected the presence of species such as ␤␥(KL/EC) 2 or ␤␥ 2 (KL/EC) 2 , of unknown biological significance. The formation of (KL/EC) 2 being unknown to us at the time, we nevertheless acquired SANS and SAXS curves of KL/EC mixtures, ␤␥ alone, and ␤␥/KL/EC. Aggregation in 42 and 100% D 2 O prevented using the differential deuteration of subunits in various H 2 O/D 2 O solutions to identify the individual components within the complex, as initially planned. A SANS experiment in water is presented here and data are summarized in Table 1. The mass of the scattering species in the KL/EC sample (55 Ϯ 8 kDa) was consistent with that of the tetramer (KL/EC) 2 (M theor ϭ 50 kDa), as expected from the SAXS results for a concentration of 270 M. The measured mass of the ␤␥-fragment (22 Ϯ 5 kDa) was also consistent with that of a monomer (M theor ϭ 20 kDa). The mean mass of the scattering species in the sample containing stoichiometric amounts of KL/EC and ␤␥-fragment (46 Ϯ 6 kDa) did not correspond to that of a complex (␤␥) 2 (KL/EC) 2 (M theor ϭ 90 kDa), nor to a noninteracting mixture of ␤␥ and (KL/EC) 2 (mean M theor ϭ 35 kDa). Instead, the measured mass was consistent with that of the ␤␥(KL/EC) complex (M theor ϭ 45 kDa), or possibly with a mixture of ␤␥(KL/EC) 2 and free ␤␥ (mean M theor ϭ 45 kDa). Analytical ultracentrifugation of the sample showed the presence of a single species (not shown), indicating that the acquired SANS curve was likely that of the ternary ␤␥(KL/EC) complex.
The masses measured by SANS suggested that the association with the ␤␥-protein could drive the dissociation of the tetramer (KL/EC) 2 into the (KL/EC) dimer. The reaction could be described by the two competing equilibria, ␤␥ ϩ (KL/EC) % ␤␥(KL/EC) with the constant K D1 , and (KL/EC) 2 % 2(KL/EC) with the constant K D2 . As we had values of the equilibrium constants K D1 ϭ 0.22 M for dissociation of the ␤␥(KL/EC) complex determined by surface plasmon resonance (see below), and K D2 comprised between 5 and 50 M for dissociation of (KL/EC) 2 estimated from SAXS mass measurements, we ran numerical simulations of the composition of the mixture at equimolar concentrations of ␤␥ and (KL/EC). The simulations showed that for concentrations greater than 100 M, the sample would contain mostly the ␤␥(KL/EC) complex, with negligible amounts of free species and (KL/EC) 2 tetramer (supplemental Fig. S5).
We therefore decided to analyze the SANS curve from the sample containing equimolar amounts (270 M) of ␤␥ and KL/EC as represented by the ␤␥(KL/EC) complex. The scattering curve I(Q) has a rather featureless profile (Fig. 4A) typical of elongated particles. The radius of gyration R g was found to be 47 Ϯ 3 Å. We have also determined the cross-sectional radius of gyration R C to be 9.7 Ϯ 0.6 Å (supplemental Fig. S6). The profile of the distance distribution function P(r) is similar to those produced by rod-like structures (45), with a steep increase at short distances and a slow decay at larger distances (Fig. 4B). The optimal D max was found to be 160 Å, which is consistent with the R g and R C values (see Equation 6 under "Experimental Procedures"). Five independent low resolution models of the ␤␥(KL/EC) complex were calculated ab initio from the SANS curve (Figs. 4C and supplemental S7). All the models are very elongated and straight with a small cross-section diameter. A common feature found in all models is an excess of mass near one end of the elongated structure. The long dimension of the excess density is about 60 Å. It is tempting to interpret this bulge as the ␤␥-protein sitting near the tip of the (KL/EC) coiled-coil dimer.
The SAXS curve of a sample containing equimolar quantities of ␤␥ and KL/EC (144 M) was very similar to the SANS curve (Fig. 4A). Because of the short exposure time due to radiation damage, quality of the SAXS data were not significantly better than that of the SANS experiments.
We also carried out rigid body modeling of the complex. Rigid body modeling seeks to minimize iteratively the difference between the experimental curve and the theoretically calculated curve of the model by translating and rotating the two partners as rigid bodies in a grid search. As rigid bodies, we used the low-resolution model of the ␤␥-fragment based on the SAXS data, and a high-resolution model of the dimerization domain of the cortexillin I from Dictyostelium discoideum as a proxy for the (KL/EC) coiled-coil dimer. The parallel homodimer coiled-coil of cortexillin comprises 200 amino acids (14 heptads) that form a long rod, which we used as a model of (KL/EC) that totals 216 amino acids and is predicted to contain at least 10 heptads plus N-and C-terminal extensions (the purification tags and the regions containing the few conserved amino acids, respectively). Fig. 4D shows five different models that fit the experimental SANS scattering curve equally well. In all models, the ␤␥-protein is positioned near the end of the elongated partner, their long axis being roughly aligned. The rigid body modeling did not allow discrimination between the two alternative orientations (head up or head down) of the ␤␥-fragment relative to the (KL/EC) rod. Manual sampling of alternative positions of the ␤␥-protein (e.g. at the extreme tip or in the middle of the rod) and orientations (e.g. perpendicular to the rod) yielded theoretical curves with worse fit to the data (supplemental Fig. S8).

Interaction between the Extracellular Part of DivIB and KL/EC Investigated by Surface Plasmon
Resonance-To evaluate the affinity of DivIB for the dimer (FtsL/DivIC) we used surface plasmon resonance (Biacore technology), where one partner (the ligand) is attached to a sensor chip, whereas the other interacting partner (the analyte) is presented over the sensor chip in a flowing solution. The surface plasmon resonance being sensitive to the mass of the particles attached to the surface of the sensor, the association kinetics can then be measured. The dissociation kinetics can be measured afterward by flowing a solution without analyte.
For the surface plasmon resonance experiments, the (KL/EC) complex was attached to the sensor chip by the N termini to present the FtsL/DivIC moieties to the solution. Attachment was achieved on Ni 2ϩ -NTA chips with both KL and EC proteins harboring an N-terminal poly-His tag. The KL/EC mixture was injected onto the sensor chip at a concentration were the (KL/EC) complex is a dimer, and not a tetramer. Association and dissociation curves could be recorded for the three soluble forms: DivIB ext , ␤, and ␤␥.
The ␤and ␤␥-truncated forms produced comparable interaction curves with (KL/EC) (not shown). We measured the kinetic parameters in the case of the ␤␥-protein.
Data are shown in Fig. 5. The theoretical curves describing the interaction according to a simple Langmuir model of association, ␤␥ ϩ (KL/EC) % ␤␥(KL/EC), were globally fit to the experimental curves.  I(Q). B, distance distribution function P(r) from SANS data. C, three-dimensional model of the ␤␥(KL/EC) complex calculated ab initio from the SANS distance distribution function. The four views correspond to 90°rotation steps around the long axis. Note that the spheres do not represent atoms or amino acids, but are dummy scattering particles. D, rigid body modeling of the ␤␥(KL/EC) complex using the lowresolution model of the ␤␥-protein obtained by SAXS (in red), and the crystal structure of cortexilin 1 (in blue) as a model for the (KL/EC) dimer. Five models were calculated by fitting their theoretical curve to the experimental data. Fits are shown below their respective models. Axes are as in A.
The extracted association rate constant was k A ϭ 7.3 ϫ 10 4 M Ϫ1 s Ϫ1 , the dissociation rate constant k D ϭ 0.016 s Ϫ1 , and the equilibrium dissociation constant was K D ϭ 220 nM, with 2 ϭ 2.26.
In the case of DivIB ext , the association curve did not reach a plateau, but continued to increase linearly, indicating aggregation (not shown). A similar linear increase was observed when DivIB ext was injected on the sensor chip devoid of (KL/EC). However, the aggregation rates on the naked surface and on the (KL/EC)-coated surface were not identical in most experiments, thus precluding subtraction. In one instance, when the linear increase was identical in the presence and absence of (KL/EC), an equilibrium constant of about 300 nM was determined, which is comparable with the constant measured with the ␤␥-protein.
The value of the affinity of the ␤␥-fragment for the (KL/EC) dimer has no direct significance for the physiological interaction between DivIB and (FtsL/DivIC), as the proteins in vivo are attached to the plasma membrane and thus restricted in two dimensions. However, the relatively high dissociation constant explains the practical difficulty in isolating the recombinant complex in vitro, as it can be partially dissociated during washing steps of co-purification procedures, or during the course of a size exclusion chromatography.
To examine the importance of the C-terminal regions of FtsL and DivIC in the interaction with DivIB, we produced truncated variants of KL and EC, termed KL* and EC*, and tested the interaction of the KL/EC* and KL*/EC dimer with the ␤␥-fragment using the surface plasmon resonance assay. The KL and EC proteins were truncated so as to preserve the predicted coiled-coil region (Fig. 6A). Fig. 6B shows the surface plasmon resonance interaction curves of the ␤␥-fragment with immobilized (KL/EC), (KL*/ EC), and (KL/EC*). Care was taken to immobilize comparable amounts of each dimer onto the sensor chip. Strikingly, truncation of either KL or EC completely abolished the interaction with the ␤␥-protein, demonstrating the importance of the C-terminal region of both FtsL and DivIC.

DISCUSSION
DivIB, FtsL, and DivIC form a ternary complex with an enigmatic function that is central to the process of bacterial division. The present work provides the first structural information on the organization of the complex, suggesting new hypotheses.
The crystal structures of the POTRA and ␤ domains from E. coli and Y. enterocolitica FtsQ have been solved (24). In contrast, our NMR and proteolysis results show that the POTRA domain from S. pneumoniae is not folded in the recombinant soluble form of the protein, as was observed with DivIB from G. stearothermophilus (26). We find it unlikely that the DivIB POTRA domain from the two latter species is truly unfolded under physiological conditions, given the conservation of the POTRA motif (25) and the results from sequence hydrophobic cluster analysis that are typical of folded proteins (11). It is more likely that the DivIB POTRA domain has a marginal conformational stability in some organisms, and that this domain is not stable in conditions that differ from those in vivo. Factors that could stabilize the structure of the POTRA domain in vivo include the pH, which may be different close to the membrane  than in the bulk water, the ionic strength, the charges at the membrane surface, or the presence of interacting protein partners. Indeed, epitopes required for the recruitment of DivIB(FtsQ) to the division site have been found in the POTRA domain, suggesting interactions with other proteins (23,24).
The low-resolution structure of the ␤-domain of pneumococcal DivIB resembles more that from E. coli FtsQ than that from G. stearothermophilus DivIB. The main reason is that the ␤-domain from G. stearothermophilus is shorter as it lacks two C-terminal ␤-strands (strands S11 and S12, Fig. 7A). Our proteolysis and NMR data concur with the crystallographic results to conclude that the ␤-domain includes S11 and S12. The absence of these ␤-strands in the ␤-domain from G. stearothermophilus is likely the unfortunate consequence of the accessible loop between S11 to the helix H5 containing a trypsin cleavage site, as proposed previously (24).
The NMR spectra of DivIB ext with and without (KL/EC) showed that the ␤-domain is involved in the interaction. The signal from the unfolded ␣and ␥-regions were not signifi-cantly affected, implying little or no interaction of these parts with the (KL/EC) dimer. However, as proton resonances of the unfolded segments are not resolved, the disappearance of some peaks from this region of the spectrum would pass unnoticed. We therefore cannot rule out completely that formation of the DivIB ext -(KL/EC) complex induces some marginal structure in the ␣or ␥-regions.
Using a deuterated ␤-domain to overcome relaxation problems resulting from the hydrodynamic properties of the ␤(KL/ EC) complex, it has been possible to detect the modification of chemical shift of 27 amide protons, which could be identified following attribution of the resonances. Given the good agreement between the low-resolution structure of the ␤-domain from pneumococcal DivIB and the crystal structure from E. coli and Y. enterocolitica FtsQ, we have tried to locate the positions affected by interaction with (KL/EC). For that purpose, we constructed an atomic model of the ␤-domain from S. pneumoniae based on that from Y. enterocolitica using the program SWISS-MODEL (46). As the sequence identity of 10% between the C, sequence alignment of the extracellular region of DivIB(FtsQ). The ␣-domain and ␥-tail are in lowercase, the ␤-domain is in uppercase. Elements of secondary structure are indicated in yellow for ␤-strands and red for ␣-helices. The lighter colors for the pneumococcal sequence indicate the hypothetical nature of the structural model. The nomenclature for the secondary structure is that from E. coli (24). Sequences were aligned using CLUSTALW (33) including sequences from additional species, in particular for the ␣-domain, and structure superimposition for the ␤-domain (PDB codes 1YR1, 2VH1, and 2VH2). Positions in the pneumococcal ␤-domain that have modified chemical shift of their amide proton upon binding of (KL/EC) are in bold.
␤-domains of both organisms is very low, the proposed model should be considered with caution and certainly not as a highresolution structure. However, the hypothetical model presented in Fig. 7, A and B, is likely sufficient to broadly locate the regions of interaction with FtsL and DivIC. Indeed, the sequence alignment between the S. pneumoniae and Y. enterocolitica proteins resulting from the model building, and the classical alignment between the S. pneumoniae and G. stearothermophilus show good agreement for the position of the secondary structure elements (Fig. 7C), lending some credence to the model.
In the model of the S. pneumoniae ␤-domain, the positions affected by the interaction with (KL/EC) are defining a large surface on the face of the bean that is formed by the ␤-sheet, whereas the opposite face formed by the ␣-helices is largely unaffected (Fig. 7, A and B). Previous mutagenesis studies in vivo have implicated the C-terminal ␤-strands of the ␤-domain in the recruitment of (FtsL/FtsB) in E. coli (24), and in the function of pneumococcal DivIB likely linked to the interaction with FtsL (17). Unfortunately, we were not able to attribute unambiguously the amide protons of this part of the sequence to NMR resonances (Fig. 2B). The absence of resonance peaks is likely due to conformational exchanges in this region of the protein. Residues 353 to 361 with attributed resonance are not included in the model due to the uncertainty of the alignment in the region. It is therefore possible that the interface with the (KL/EC) dimer extends to the C-terminal region at the "top" of the bean.
In the absence of the high resolution structure of the pneumococcal ␤-domain, it is premature to test the importance of residues at the interface with FtsL and DivIC by site-directed mutagenesis. As the interface covers a large part of one face of the ␤-domain, it is possible that single mutations would not be sufficient to suppress the interaction, although D237N and A252P substitutions in E. coli FtsQ each abolished recruitment of FtsL to the division site (24,47).
The ab initio model of the ␤␥(KL/EC) complex based on the SANS data, corroborated by rigid body modeling, indicates that the ␤-domain is positioned near the tip of the rod constituted by the long (KL/EC) coiled-coil. Previous controls had shown that DivIB ext does not interact with the artificial K/E part (11). It was therefore reasonable to assume that the ␤-domain binds to the C-terminal regions of FtsL and DivIC. This result was confirmed by the complete loss of the interaction measured by surface plasmon resonance when either KL or EC were truncated of their C termini.
When viewing the models presented in Fig. 4, it is to be remembered that a large part of the long rod is constituted by the k5 and e5 coiled-coil moieties that were added to constrain the dimerization of the extracellular regions of FtsL and DivIC (11). The purification tags of the KL and EC proteins are also present and contribute to the length of the complex. However, the five predicted heptads of the extracellular coiled-coil of the FtsL/DivIC dimer should extend about 50 Å from the membrane to present the C-terminal regions that contain the conserved residues. The POTRA or ␣-domain of E. coli FtsQ is about 30 Å long and there are about eight additional residues between the transmembrane segment and the POTRA domain, so that the lower part of the ␤-domain could stand about 30 to 60 Å from the membrane, whether the eight additional amino acids are tightly folded or fully extended. However, the sequence of the eight residues proximal to the membrane does not have the polar character to constitute a long solvent-exposed extended linker. If the base of the ␤-domain does not reach 50 Å from the membrane, the (FtsL/DivIC) rod may need to be tilted to present its extremity to the ␤-domain. An angle of 37°would allow the end of the FtsL/DivIC coiled-coil to make contact with the base of a ␤-domain 40 Å from the membrane. Note that tilt angles of this magnitude are not unheard of for transmembrane helices (48). Alternatively, the presence of a kink between the transmembrane segment and the extracellular coiled-coil of FtsL and DivIC cannot be excluded. Indeed, the prediction score for a helical structure, using various programs, falls in the segment of a few residues next to the membrane (Fig. 8). Several genetic studies have failed to indicate a role of the POTRA domain in the recruitment of FtsL and DivIC(FtsB). Also, our NMR data do not support an interaction between the POTRA domain and the (FtsL/DivIC) dimer. If the C-terminal segments of the dimer constitute the sole region that interacts with the ␤-domain of DivIC(FtsB), the resulting model leaves the coiled-coil rod free for interacting with other proteins. In this respect, it is striking that the Campylobacterales, which divide without FtsQ, have retained the coiled-coil part of FtsL and FtsB. The absence of the C-terminal regions of FtsL and FtsB in Campylobacterales indicates that the coiled-coil region plays a role per se, and not simply as a way to bring the C-terminal parts together. The coiled-coil rod of the FtsL/ DivIC(FtsB) dimer, for example, could interact with the septal penicillin-binding protein FtsI, to regulate the function of this enzyme in the assembly of the cell wall. Indeed, a dimer of FtsL and a C-terminal truncated form of FtsB is still able to recruit FtsI to the division site (27).
It is paradoxical that the coiled-coil part of the FtsL/ DivIC(FtsB) dimer appears to be the essential feature of this complex as it exhibits no sequence conservation. This paradox prompts us to propose another speculative function for the dimer. One of the many processes coordinated during bacterial cell division is the membrane invagination. It is possible that the onset of the invagination, which implies a sharp membrane curvature, might require or benefit from destabilization or modification of the lipid bilayer properties. In this respect, it is noteworthy that oligomerization of bitopic membrane proteins, driven by the formation of coiled-coils, is at the core of eukaryotic and viral membrane fusion events (49). Most interestingly, it was found in biophysical studies that oligomerization of a transmembrane fusogenic peptide, constrained by an artificial extramembranous coiled-coil domain, modifies and destabilizes the membrane (31). An analogous phenomenon could occur during bacterial cell division, where the dimerization of FtsL and DivIC(FtsB) might bring their transmembrane segments in close proximity, thereby inducing local modification of the membrane properties, which could be important for the invagination.
In these hypothetical models of the function of FtsL and DivIC, the capping provided by DivIB may simply stabilize the coiled-coil. This could explain why DivIB is essential only at high temperatures in B. subtilis (8,19), if the (FtsL/DivIC) dimer of this species were stable enough at low temperature. This model could also explain why Campylobacterales can divide without FtsQ, if the interaction between their FtsL and FstB were strong enough. We propose to reverse our way of considering the interaction between FtsL and DivIC(FtsB). Instead of viewing the transmembrane segments and coiledcoil regions as a means of bringing together the C-terminal parts, the interaction of the C-terminal parts of FtsL and DivIC(FtsB) with DivIB(FtsQ) could help to zip together the coiled-coil helices and bring the transmembrane segments together.
We have seen no structuration of the ␣-domain upon binding of (KL/EC), which suggests the absence of interaction of the POTRA domain with the (FtsL/DivIC) dimer. No mutation affecting the recruitment of FtsL/DivIC(FtsB) to the division site was found in the POTRA domain of E. coli FtsQ (24). Also, the complete POTRA domain can be deleted from B. subtilis DivIB (23) without compromising growth. In contrast, mutations in the POTRA domains have been found to affect the recruitment of FtsQ and DivIB to the division site of E. coli and B. subtilis (23,24). Could the POTRA domain intervene in the function of FtsL and DivIC(FtsB)? POTRA domain stands for polypeptide-transport-associated domain (25). Apart from DivIB(FtsQ), POTRA domains are found in ␤-barrel outer membrane proteins of Gram-negative bacteria, where they are thought to assist the folding of ␤-barrels by binding the unfolded polypeptide chain in an extended conformation (32). Coiled-coils, such as that found in FtsL and DivIC(FtsB), are assemblies of ␣-helices that are usually not folded individually prior to assembly. Thus, the extracellular region of FtsL and DivIC(FtsB) are likely to be unfolded without their partners (28). The POTRA domain could offer a binding interface, much as a chaperone, to FtsL and/or DivIC prior to their association. FtsL would be the preferred substrate of this chaperoning function of the POTRA domain, as it appears in two-hybrid experiments that FtsL alone interacts with FtsQ, whereas FtsB requires the presence of FtsL to interact with FtsQ (13). In these experiments, the interaction of FtsL might involve the POTRA domain, whereas the (FtsL/FtsB) dimer would interact with the ␤-domain of FtsQ.
The interaction of the ␤-domain of DivIB with the C-terminal regions of the FtsL/DivIC dimer revealed by our structural and interaction studies in vitro, opens new directions of investigation by suggesting that the coiled-coil region of FtsL and DivIC is free to interact with other partners, or plays a central role in itself.