Common Evolutionary Origin for the Rotor Domain of Rotary Atpases and Flagellar Protein Export Apparatus

The V1- and F1- rotary ATPases contain a rotor that rotates against a catalytic A3B3 or α3β3 stator. The rotor F1-γ or V1-DF is composed of both anti-parallel coiled coil and globular-loop parts. The bacterial flagellar type III export apparatus contains a V1/F1-like ATPase ring structure composed of FliI6 homo-hexamer and FliJ which adopts an anti-parallel coiled coil structure without the globular-loop part. Here we report that FliJ of Salmonella enterica serovar Typhimurium shows a rotor like function in Thermus thermophilus A3B3 based on both biochemical and structural analysis. Single molecular analysis indicates that an anti-parallel coiled-coil structure protein (FliJ structure protein) functions as a rotor in A3B3. A rotary ATPase possessing an F1-γ-like protein generated by fusion of the D and F subunits of V1 rotates, suggesting F1-γ could be the result of a fusion of the genes encoding two separate rotor subunits. Together with sequence comparison among the globular part proteins, the data strongly suggest that the rotor domains of the rotary ATPases and the flagellar export apparatus share a common evolutionary origin.


Introduction
Two types of rotary ATPases are found in biological membranes; V o V 1 (V type ATPase) and F o F 1 (F type ATPase) [1][2][3][4]. Evolutionary counterparts of the eukaryotic V o V 1 are found in most archea and some bacteria (often referred to as A-type ATPase). Both V o V 1 and F o F 1 couple ATP synthesis and hydrolysis to proton translocation across the membrane by rotation of the rotor apparatus against the surrounding stator which includes the catalytic A 3 B 3 or a 3 b 3 hexamer (Fig. 1a). Sequence and structural comparison of the two ATPases indicate significant homology between the catalytic subunits, but not between the subunits of the central rotor domain [5]. For example, V o V 1 lacks a counterpart of the rotor shaft F 1 -c subunit [1,4,5].
The minimal ATP-driven rotary unit of F o F 1 is F 1 , which is comprised of three different proteins with a stoichiometry of a 3 b 3 c. The F 1 -c contains two distinct domains; a coiled-coil domain that penetrates the central cavity of the a 3 b 3 cylinder from the top to the bottom, and a globular domain containing a a/b fold which makes contact with the F 1 -e (Figs. 1a,b).
For V o V 1 , the minimal rotary unit consists of the A 3 B 3 D subunits [6]. In addition, analysis of the secondary structure of the D subunit predicts the presence of long a-helices at both the amino and carboxyl termini as found in the F 1 -c (Fig. 1b, Supplementary  Fig. S1a). On the basis of this, it has been suggested the D subunit is likely to be a structural and mechanistic analog of F 1 -c despite the lack of any significant sequence similarity [5]. However, F o F 1 lacks a counterpart of the V 1 -F subunit [1,4,5], which has a typical globular a/b fold [7] (Fig. 1c). Intriguingly, the X-ray structure of the entire V 1 of T. thermophilus revealed that the central rotor subunit contains both the coiled-coil (D) and globular domains (F), suggesting that the V 1 -F and V 1 -D subunits together form the counterpart of the a/b domain of F 1 -c [8].
Mulkidjanian et al. have proposed a scenario to explain the origin of rotary ATPases, where rotary ATPases share an evolutionary origin with the bacterial flagellar and non-flagellar type III export systems [5]. The flagellar export apparatus consists of a membrane-embedded export gate composed of FlhA, FlhB, FliO, FliP, FliQ and FliR, and a water-soluble ATPase complex consisting of FliH, FliI, and FliJ (Fig. 1a, lower panel) [9]. Components of the flagellar ATPase complex, which allows the export gate to efficiently utilize proton motive force across the cytoplasmic membrane as an energy souse for protein translocation [10,11], exhibit extensive structural and sequence similarity to catalytic and rotor subunits of rotary ATPases. For instance, the atomic structure of FliI ATPase of the flagellar export apparatus is remarkably similar to the F 1 b/a and V 1 A/B subunits [12]. FliJ, which is a soluble export component protein, also shows a striking structural similarity to the coiled-coil region of F 1 -c [13]. This structural motif is also seen in the YscO-like protein, CT670 from Chlamydia tracomatis, a FliJ homolog of the non-flagellar type III export apparatus [14]. Interestingly, FliJ promotes the formation of a homo-hexamer of FliI by binding to the center of the ring, facilitating the FliI ATPase activity [9]. A very similar arrangement is observed in the F o F 1 -ATPase with the antiparallel ahelical coiled coil formed by the N-and C-terminal regions of the c subunit penetrating into the central cavity of the a 3 b 3 ring. These findings indicate that the type III export system has a F 1 -or V 1 -like structure and share a common evolutionary origin.
Here we aimed to explore the evolutionary relationship between the rotor domains of the V o V 1 and F o F 1 and the type III export system. FliJ formed chimeric complex with A 3 B 3 of V o V 1 and promoted their ATPase activity. Single molecular analysis indicates anti-parallel coiled coil structure (FliJ structure) protein functions as a rotor axis. These results strongly suggest that the FliJ structure proteins are the ancestral subunit of the rotor subunit of rotary ATPases. We also discuss the evolutionary relationship between globular domain of F 1 -c, and V 1 -F.
FliJ was expressed in BL21(DE3)pLysS and purified as described previously [13]. FliJ(C32T/I61C/I67C) was created by QuikChange mutagenesis (Stratagene), expressed in BL21(DE3)pLysS, biotinylated and purified as the same protocol used for FliJ(C32T/I67C) [13]. A 3 B 3 was expressed and purified as previously described [15]. The A 3 B 3 -FliJ complex was reconstructed from the purified FliJ and A 3 B 3 . FliJ and A 3 B 3 were mixed and incubated overnight at 23uC. The mixture was applied to a HilLoad Superdex 200 column in 20 mM MOPS-NaOH pH 7.0 and 100 mM NaCl. The fraction containing the A 3 B 3 -FliJ complex was pooled and stored for future use.

Rotation experiments
Biotinylated FliJ(C32T/I61C/I67C) and His-tagged A 3 B 3 were mixed and incubated overnight at 23uC. The A 3 B 3 -FliJ mixture was applied to 5-ml polypropylene column containing Ni-NTA agarose (Qiagen), and A 3 B 3 -FliJ complex was eluted with a sodium phosphate buffer (pH 7.2) containing 200 mM imidazole and 300 mM NaCl. BSA buffer, which is buffer A (50 mM Tris-Cl, pH 8.0, 100 mM KCl, 2 mM MgCl 2 ) containing 2 mg ml 21 BSA, was infused into the flow cell to prevent nonspecific binding, and then the A 3 B 3 -FliJ complexes diluted 1:10 into BSA buffer were attached on the Ni-NTA glass and incubated for 15 minutes at 23uC. Unbound molecules were washed out with BSA buffer, and streptavidin-coated gold spheres with a diameter of 40-nm (BioAssay Works) diluted into BSA buffer and infused into the flow cell. After incubation for 15 minutes at 23uC, unbound gold spheres were washed out with buffer A. After infusion of buffer A containing 4 mM Mg-ATP, 2.5 mM phosphoenol pyruvate, and 0.5 mg ml 21 pyruvate kinase, rotation of the gold spheres were observed by a dark-field microscope (BX53, U-DCW, UPlanFLN 1006, PE 56; Olympus) and recorded with a high-speed camera (ICL-B0620M-KC, IMPERX) at 0.8-1.6-ms intervals at 23uC. For the rotation assay of A 3 B 3 DF f or A 3 B 3 JL, the biotinylated enzyme (1-5 nM) in buffer A was applied to the flow cell and incubated for a few minutes at 23uC. Streptavidin-coated magnetic beads (100-300 nm) and Ni 2+ -NTA coated cover glasses were prepared as previously described [20,21]. Unbound enzyme was washed out with 20 ml of buffer A. Then, 20 ml of buffer A with 2 mg ml 21 BSA was infused to the flow cell and incubated for ,30 s to prevent nonspecific binding. The BSA solution in the chamber was washed out with 20 ml of buffer A. Then, buffer A containing streptavidin coated magnetic beads (10 10 ,10 11 particles ml 21 ) were infused into the flow cell and incubated for a few minutes. Unbound beads were washed out with 20 ml of buffer A. After infusion of 80 ml of buffer A containing Mg-ATP at the indicated concentration, 2 mM MgCl 2 , 2.5 mM phosphoenol pyruvate, and 0.5 mg ml 21 pyruvate kinase, rotation of the bead was recorded with a high speed camera (Eclips, IN) at 1000 frames per second (f.p.s.) using a phase-contrast microscope (IX70, Olympus) with 6100 objective lens (N.A., 1.30, Olympus) at 23uC. Images were captured as an 8-bit AVI file. The centroid of the bead images was calculated [20,21].

Electron cryo-microscopy and image analysis
Sample grids were prepared by applying 3 ml of a protein solution containing the in vitro reconstructed A 3 B 3 J rings (150 mg/ ml) onto a holey carbon grid (Quantifoil R0.6/1.3, Quantifoil Micro Tools, Jena, Germany), which had been glow discharge for 20 s before use. The grids were blotted onto filter paper for 5 s to remove excess solution, vitrified in liquid ethane at 2196uC using a Vitrobot (FEI, Eindhoven, Netherlands) and transferred into a liquid nitrogen storage capsule. Particle images were recorded with a 4 K64 K Slow Scan CCD camera (TemCam-F415MP, TVIPS) mounted on a JEM-3200FSC electron microscope (JEOL, Tokyo, Japan), equipped with a liquid-helium cooled specimen stage, a Vtype energy filter and a field-emission electron gun operated at an accelerating voltage of 200 kV. Electron micrographs were collected at 50 K with a magnification of 6140,000, corresponding to 1.07 Å /pixel. Focal pairs of the micrographs were recorded at a defocus between 1.5 and 2.5 mm for the first micrograph and at a defocus between 3.5 and 5.5 mm for the second. Electron dose was set to 30 e 2 /Å 2 for both micrographs. The particle images were processed using the EMAN software package [22]. The focal pairs were merged using FOCALPAIR [22,23]. The defocus amplitude, envelope and noise values were determined using CTFIT [22]. Micrographs showing significant astigmatism or drift were discarded. The particle images were selected with BOXER [22] and the boxed particle images were aligned and classified using REFINE2D.PY [22].

Phylogenetic tree analysis
All sequences used in this study were aligned using the MAFFT program (http://mafft.cbrc.jp/alignment/server/) [24]. Phylogenetic and molecular evolutionary analyses were conducted using MEGA version 5 [25]. The phylogenetic tree was constructed using the Maximum-Likelihood (ML) method under the Jones-Taylor-Thornton model by using MEGA.

Reconstitution of A 3 B 3 J
Structural sequence alignment has previously revealed that FliJ shares some amino acid conservation with F 1 -c [13]. The structural and sequence similarity suggests that FliJ may behave as a counterpart of rotor subunit of rotary ATPases. To test this possibility, we investigated whether FliJ binds to the center of the A 3 B 3 ring of T. thermophilus, forming the A 3 B 3 J complex. A 3 B 3 was mixed with excess amounts of FliJ, and the mixture incubated overnight at room temperature. The mixture was applied to a gel filtration column to remove free FliJ. The retention time of the complex peak was almost identical to that of A 3 B 3 alone (Fig. S2a). SDS-PAGE analysis revealed that FliJ co-eluted with A 3 B 3 , indicating the formation of the A 3 B 3 J complex (Fig. 2a). Previous studies have shown that V 1 and A 3 B 3 D are resistant to ATPinduced dissociation [15] while A 3 B 3 completely dissociates into monomeric A and B subunits by incubation with 1 mM ATP for 30 min (Fig. 2b). In contrast, the A 3 B 3 J complex is at least partially resistant to ATP-induced dissociation (Fig. 2b), indicating that the presence of FliJ stabilizes the A 3 B 3 complex to a significant degree. ATP hydrolysis by V 1 or A 3 B 3 D proceeds at a steady rate for a few minutes, then decelerates slowly, due to ADP inhibition [16]. In contrast, the ATPase activity of A 3 B 3 decelerates rapidly, reaching a steady state rate within a few minutes owing to rapid ATPinduced dissociation of A 3 B 3 [16]. In the case of A 3 B 3 J, the ATP hydrolysis profile is similar to that of V 1 or A 3 B 3 D exhibiting continuous ATP hydrolysis activity after an initial burst phase (Fig. 2c). The turnover rate of A 3 B 3 J is ,7.0 s 21 , almost twice that of A 3 B 3 (,4.0 s 21 ). Together, these results suggest that FliJ stabilizes the A 3 B 3 hexamer by binding to A 3 B 3 and promotes continuous ATPase activity in a similar manner to the V 1 -D subunit.

Electron cryo-microscopy analysis of A 3 B 3 J
In order to identify the location of FliJ in the A 3 B 3 J complex, we analyzed the images of frozen-hydrated A 3 B 3 J and A 3 B 3 particles embedded in vitreous ice by electron cryo-microscopy. Since the distribution of the particle orientation was strongly biased to endon view, we aligned and averaged the end-on images of each particle. The averaged image of A 3 B 3 shows a hetero-hexameric ring structure with an unoccupied central hole of 2 nm in diameter (Fig. 3a), consistent with the crystal structure of A 3 B 3 [17]. In contrast, the averaged image of A 3 B 3 J, clearly shows extra density in the interior of the central hole but located off center, close to one of the peripheral subunits (Fig. 3b). These observations strongly suggest that FliJ penetrates into the central hole of the A 3 B 3 complex in a way similar to the D-subunit of V 1 -ATPase [8].

Rotary motion of FliJ in A 3 B 3 without uni-directionality
We attempted to demonstrate the rotary motion of FliJ against A 3 B 3 immobilizing A 3 B 3 J on a Ni-NTA-coated glass surface through His 10 -tags introduced at the N terminus of the A subunits, and attaching a 40-nm streptavidin-coated gold colloid (40-nm bead) to the biotin-labelled FliJ (Fig. 4a). Beads were imaged by dark-field microscopy, and beads motions were recorded on a fastframing CCD camera at speeds up to 1,000 frames s 21 . We found rotating beads attached to the FliJ in A 3 B 3 . It was confirmed that gold beads observed under the microscope were attached to the A 3 B 3 J complex through FliJ because very few beads were found without A 3 B 3 J. Two to ten beads showing rotary motion were usually found in a single flow cell. However, they did not show apparent uni-directionality. For instance, both rotating beads showing clockwise or anti clockwise rotation (Fig. 4 b-e) were found. Some beads showed stepwise rotary motion (Fig. 4 f,g). The rotary motion of A 3 B 3 J was also observed without ATP in the infusion buffer but the number of beads showing rotary motion apparently decreased; one to three beads showing rotary motion were found in a single flow cell. These results indicate that FliJ in the A 3 B 3 does not function as a perfect rotor axis. FliJ is only weakly bound to A 3 B 3 , thereby dissociating FliJ from the complex after successive gel filtration chromatography of the A 3 B 3 J complex (Fig. S2b). In agreement with this, the ATPase activity of A 3 B 3 J (7.0 s 21 ), although significantly higher than A 3 B 3 , is much lower than that of V 1 (65 s 21 ) or A 3 B 3 D (25 s 21 ), reflecting the relative instability of the A 3 B 3 J complex. Therefore, the low binding affinity of the FliJ for A 3 B 3 is likely to reduce the chance of FliJ rotation being coupled with a conformational change of A 3 B 3 by ATP hydrolysis.
Anti-parallel coiled coil structure (FliJ structure) functions as a rotor FliJ adopts an anti-parallel coiled coil structure composed of both N-and C-terminal a-helices and is similar to that of V 1 -D or F 1 -c [13]. If we assume that the rotor subunits of the rotary ATPases and the flagellar type III export apparatus share a common evolutionary origin, then the ancestral rotor subunits  adopt an anti-parallel coiled-coil structure, that is, the FliJ structure. In order to demonstrate that the FliJ structure functions as a rotor, we constructed a gene encoding a FliJ structure (FliJ-like protein, termed JL hereafter). The JL gene encodes an antiparallel coiled coil region of V 1 -D composed of both N-and Cterminal a-helices, without the additional region (52 a.a. length) between the N-and C-terminal helices (Fig. S1a). Then we constructed an expression vector encoding the JL gene along with atpA and atpB genes (Fig. S3a). The JL protein was co-expressed along with the A and B subunits in Escherichia coli and purified the A 3 B 3 JL complex to homogeneity (Fig. 2a). The A 3 B 3 JL was active as an ATPase and the V max value of the A 3 B 3 JL is similar to that of A 3 B 3 D (Fig. S3b). Using a direct observation system similar to that described for V 1 [6], the rotation of the A 3 B 3 JL was able to be visualized (Fig. 5a). We observed rotating beads attached to the A 3 B 3 JL when the flow cell was infused with buffer containing 4 mM or 10 mM ATP (hereafter [ATP]) (Fig. 5b). The results clearly indicate that the JL protein functions as a rotor. At 10 mM [ATP], A 3 B 3 JL showed a clear stepwise rotation, pausing every 120u like F 1 or V 1 . Interestingly, frequent backward steps were observed in rotation of A 3 B 3 JL but were seldom seen in stepwise rotation of either V 1 or F 1 , indicating that the extra 53 amino acid residues inserted between the two helices were necessary for continuous rotation without irregular motions (Fig. 5b, inset). Taken together, we propose that an anti-parallel coiled-coil fold unit like FliJ is sufficient to function as a rotor for the ATP-driven rotary motor. F 1 -c like rotor derived from V 1 -D and V 1 -F As shown in Figure 1, the similar folds of V 1 -F and the globular domain of F 1 -c prompted us to probe an evolutionary relationship between these two proteins, i.e if could F 1 -c be derived from two separate proteins as observed for V 1 -D and V 1 -F. To investigate the functional relationship between V 1 -F and globular domain of F 1 -c, we constructed an expression vector of a mutant V 1 containing a fusion of the genes coding for subunits D and F. The gene coding for F (atpF) was inserted between the two helical regions of the gene coding for D (atpD; see Fig. S4a) and the resulting fusion protein was termed DF f . The mutant ATPase containing the fused protein (A 3 B 3 DF f ) was expressed in E. coli and purified to homogeneity (Fig. 2a). The A 3 B 3 DF f showed an ATPase activity and exhibited simple Michaelis-Menten kinetics, nearly equal to those of wild type V 1 , which contains separate D and F subunits in the rotor [18]. Using a direct observation system similar to that described for V 1 , rotation was visualized via a bead attached obliquely to the helical region in the DF f (Fig. 5c). We  Table S1. doi:10.1371/journal.pone.0064695.g006 observed rotating beads attached to the A 3 B 3 DF f at 4 mM or 10 mM ATP (Fig. 5d). Stepwise rotation of the A 3 B 3 DF f pausing every 120u was also observed at 10 mM ATP. In this case the frequent backwards steps seen for A 3 B 3 JL were not observed (Figs. 5d, inset). Together, the identical rotation behavior and kinetic parameters of A 3 B 3 DF f confirm that the DF f fully functions as a shaft in the rotary motor. These observations indicate that the V 1 -F when fused to V 1 -D, functions in the rotor in the same way as the globular domain of F 1 -c and suggest an evolutionary relationship between V 1 -F and V 1 -D and F 1 -c. In other words, F 1c could be the product of fusion of the genes encoding an ancestral helical protein similar to V 1 -D and an ancestral globular protein similar to V 1 -F.

Discussion
In this study, we have provided several lines of evidence on the functional similarity of FliJ of the flagellar type III export system to the rotor subunit in rotary ATPases. Analysis of reconstituted A 3 B 3 J revealed that FliJ stabilizes the A 3 B 3 hexamer by penetrating into A 3 B 3 and promotes continuous ATPase activity in a manner similar to the V 1 -D subunit. Although FliJ in A 3 B 3 does not show unidirectional rotation coupling with ATP hydrolysis due to its low binding affinity for A 3 B 3 , The JL protein is sufficient for functioning as a rotor of the ATP-driven rotary motor. Based on very recent high resolution X-ray crystal structure of V 1 from E. hirae, Arai and co-workers have proposed that the interaction of the coiled-coil part of V 1 -D with the A subunit is essential for rotation of V 1 -D against A 3 B 3 heterohexamer [26]. In their V 1 structure, the globular-loop part of V 1 -DF is rarely in contact with the A 3 B 3 hexamer. Our results, in which the anti-parallel coiled-coil structure is sufficient as a rotor in rotary ATPases, are consistent with their structural study.
Here we propose that V 1 -D, F 1 -c and FliJ evolved from a common evolutionary origin. The JL protein derived from V 1 -D functioned as a rotor shaft ( Figure 5A), strongly suggesting that FliJ maintains features of a prototype rotor. Because the rotor is composed of separate helical and globular subunits in V o V 1 , this rotor is an intermediate between the ancestral rotor and the F 1 -c, which is a single protein containing both helical and globular domains. Therefore, we propose a possible scenario of evolutional process of rotor apparatus in rotary motors (Fig. 6). In this scenario, F 1 -c evolved from a gene fusion of genes encoding the ancestral V 1 -D and V 1 -F like proteins. However, because there is the sequence and structural diversity between V 1 -F and in the globular domain of F 1 -c. we cannot exclude the possibility that V 1 -F is not ancestral gene of globular domain of F 1 -c. V 1 -F is structurally similar to CheY, a regulatory subunit of the bacterial flagellar motor, which functions to switch the direction of rotation [7]. A phylogenetic tree analysis using the Maximum Likehood (ML) method indicated that V 1 -F is evolutionarily related to CheY rather than to the globular domain of F 1 -c (Fig.  S5). The topology of V 1 -F is also more similar to that of CheY. (Fig. S5). This indicates that V 1 -F and CheY share a common evolutionary origin. In contrast, the sequence and structural diversity between V 1 -F/CheY and in the globular domain of F 1 -c can be explained. The central rotor apparatus of the two rotary ATPases contain significant structural differences. The V o sector contains the funnel shaped C subunit (V o -d subunit), which serves as a socket for the DF rotor in V 1 [1,19]. In contrast, F 1 -c attaches directly onto the F o -c ring while the F 1 -e forms contacts with both the F 1 -c and F o -c ring [1,3] (see Fig. 1a). It is possible that the differences in contact features between V 1 -F and the globular domain of F 1 -c have promoted structural and sequence diversity of the rotors during the evolution of the two different ATPases.
In contrast to V 1 -DF and the F 1 -c, there is little similarity between the other central rotor domain of the V o V 1 -and F o F 1 (see Fig. 1a). The F 1 -e, composed of an N-terminal b sandwich and a short C terminal helix [1,3] shows neither sequence nor structural similarity to the equivalent V o -d (prokaryotic C subunit). Assuming that V o V 1 have conserved the ancestral form in the rotor apparatus, V o -d has been replaced with F 1 -e during evolution of F o F 1 .