Co-folding of a FliF-FliG split domain forms the basis of the MS:C ring interface within the bacterial flagellar motor

Summary

The interface between the membrane (MS) and cytoplasmic (C) rings of the bacterial flagellar motor couples torque generation to rotation within the membrane. The structure of the C-terminal helices of the integral membrane protein FliF (FliF_C) bound to the N-terminus of the switch complex protein FliG (FliG_N) reveals that FliG_N folds around FliF_C to produce a topology that closely resembles both the middle and C-terminal domains of FliG. The interface is consistent with solution-state NMR, SAXS, in vivo interaction studies and cellular motility assays. Co-folding with FliF_C induces substantial conformational changes in FliG_N, and suggests that FliF and FliG have the same stoichiometry within the rotor. Modeling the FliF_C:FliG_N complex into cryoEM rotor density updates the architecture of the middle and upper switch complex and shows how domain shuffling of a conserved interaction module anchors the cytoplasmic rotor to the membrane.

eTOC Blurb

FliF and FliG comprise the MS-ring and upper C-ring of the bacterial flagellar motor. Lynch et al. use X-ray crystallography, SAXS, NMR, and in vivo studies to reveal how FliF:FliG fold into a single domain, whose toplogy is found elsewhere in FliG, and generate an updated model of the upper flagellar rotor.

Introduction

The bacterial flagellar motor is the principle organelle that enables motile bacteria to move within their environment. The motor can be divided into three major components: (1) the filament, which serves as the propeller of the motor, (2) the hook, a flexible adaptor between the filament and cell body, and (3) the basal body, which generates torque and induces flagellar rotation switching. The basal body (Figure 1) is comprised of a series of transmembrane rings that enclose a central rod. Powered by the proton or sodium gradient that spans the inner membrane, torque is generated by interactions between the membrane-embedded ion channels and the switch complex located in the cytosolic space (Berg, 2003; Minamino et al., 2008, 2015; Sowa, 2008; Chen et al., 2011) The switch complex forms the C-ring and is composed of the proteins FliG, FliM, and FliN (or FliY) (Figure 1). The switch complex rotates either counterclockwise (CCW) or clockwise (CW) and in doing so dictates whether the cell swims smoothly or tumbles in solution. Interaction of the switch complex with the phosphorylated form of the response regulator CheY causes switching of rotation direction (Berg, 2003; Sowa, 2008).

Figure 1.

Schematic diagram of the bacterial flagellar motor.

A prerequisite for motor operation and assembly during the early stages of flagellar biosynthesis is the correct positioning of the switch complex relative to the MS-ring (Macnab, 2003; Grünenfelder et al., 2003; Chevance et al., 2008; Minamino et al., 2008, 2015). Located in the inner membrane and composed of approximately 25 copies of a single transmembrane protein, FliF, the MS-ring is the first circular structure to form during flagellar assembly (Figure 1) (Macnab, 2003; Chevance et al., 2008; Sowa, 2008). The MS-ring adheres to the switch complex through interactions mediated between the C-terminal tail of FliF (FliF_C) and the N-terminal region of FliG (FliG_N) (Francis et al., 1992, Thomas et al., 2001; Brown et al., 2002; Grünenfelder et al., 2003). Electron microscopy, site-directed alanine-scanning mutagenesis, and fusion/deletion studies have revealed that these two proteins directly interact with one another in a 1:1 stoichiometric ratio (Thomas et al., 1999, 2001, 2006; Ogawa et al., 2015). The structure of FliG has been determined for the full-length protein (Lee et al., 2010), and the individual domains (Brown et al., 2002, Minamino et al. 2011), alone and in complex with domains of FliM (Paul et al., 2011; Vartanian et al., 2012; Lam et al., 2013; Sircar et al., 2015). FliF is an integral membrane protein with extensive periplasmic domains homologous to the injectosome SctJ/D proteins and the sporulation factors SpoIIIAG/H (Bergeron, 2016). FliG and FliF orthologs are found in essentially all flagellated (and some non-flagellated) bacteria and thus this interaction can be considered a universal contact made in bacterial flagella (Levenson et al., 2012; Bergeron, 2016).

To characterize the interaction between FliF_C and FliG_N, Levenson et al. employed tryptophan fluorescence and ¹H-¹⁵N TROSY-HSQC nuclear magnetic resonance spectroscopy to map the binding surface of T. maritima FliF_C on FliG_N (Levenson et al., 2012). The K_d between FliG_N and FliF_C was measured in the low nanomolar range (< 40 nM) and FliF_C binding caused chemical shift perturbations of backbone resonances across the entirety of FliG_N. As a whole, the data suggested that the FliG_N domain orders as a result of FliF_C binding – an effect consistent with multi-angle light scattering experiments that demonstrated the dissociation of FliG homodimers to form mixed FliF_C:FliG heterodimers upon FliF_C addition (Levenson et al., 2012). A strong interaction between FliF_C and FliG_N agrees well with previous observations made when purifying intact flagellar motors from Salmonella cells. Intact basal bodies readily lose the FliM and FliN regions of the C-ring; however, very low pH is required to induce dissociation of the FliF:FliG interface, owing to the unique strength of the interaction (Francis et al., 1994). More recently, a similar interaction between FliF and FliG was reconstituted with a soluble full-length form of FliF from a marine Vibro species (Ogawa et al., 2015).

Herein, we report the crystal structure of a complex between FliF_495–532 (FliF_C) and FliG_1–98 (FliG_N). All three domains of full-length FliG have a strikingly similar topology when FliG_N is bound to FliF_C. Indeed, FliG_N folds around the FliF_C helices and produces a structure similar to that of the FliG middle domain (FliG_M) and the FliG C-terminal domain (FliG_C) (Brown et al, 2002; Lee et al., 2010). Solution NMR, small-angle X-ray scattering (SAXS), and biochemical data are fully consistent with the structure, and help differentiate between two alternative complexes found in the crystal lattice. The FliF_C:FliG_N structure was then used to model the MS/C-ring interface in the context of cryo-EM density for whole Salmonella rotors (Thomas et al., 2006). An extended C-ring structure was generated taking into account previously placed FliG_MC:FliM_M units (Sircar et al., 2015). Targeted cross-linking and motility assays of structure-guided variant proteins reveal the functional importance of the FliF_C:FliG_N interaction

Results

Solution-state properties of an engineered FliF_C:FliG_N fusion protein

FliF_C:FliG_N {FliF_C(495–532):FliG_N(1–98) for crystallography and FliF_C(495–532):FliG_N(1–102) for NMR} were produced as chimeric fusion proteins to ensure a 1:1 component ratio (SI Figure 1). The constructs were engineered such that FliG and FliF are linked together via a sequence containing a His₈ tag bracketed by two TEV protease sites (or one TEV protease site between His₈ and FliG_N for the NMR studies). ¹H-¹⁵N TROSY-HSQC NMR spectroscopy was employed to evaluate the resulting complex before and after TEV proteolysis (Figure 2B). Additionally, the cleaved complex was compared to that of ¹⁵N labeled FliG_N incubated with a synthetic FliF_C peptide corresponding to the identical sequence encoded by the fusion construct (Figure 2A) (Levenson et al., 2012) The ¹H-¹⁵N TROSY-HSQC spectra of the FliF_C:FliG_N fusion construct after TEV proteolysis (Figure 2A, red) is nearly identical to that of ¹⁵N labeled FliG_N incubated with stoichiometric amounts of the corresponding FliF_C peptide (Figure 2A, black). Thus, the cleaved fusion protein has a conformation similar to that of FliG_N mixed with FliF_C. Similarly, there are little differences in the measured ¹H-¹⁵N resonances before (black) and after (red) TEV proteolysis (Figure 2B) indicating that the intact and cleaved proteins are remarkably similar in organization. Resonances of isotopically labeled TEV-proteolyzed FliF_495–532:FliG_1–102 complex were assigned using standard 3D NMR techniques, yielding assignments for 93% of residues (SI Figure 2A). Cα chemical shift deviations from random coil values showed both FliF_495–532 and FliG_1–102 within the complex to be well-folded, with alpha helical structure (SI Figure 2B). Overall, the data indicate the FliF_C:FliG_N complex is the same whether produced as a fusion protein or as separate components and that the presence of a linker does not prevent complex assembly.

Figure 2.

NMR Characterization of FliF_495–532:FliG_1–102 complex. (A) Black: TROSY-HSQC data of ¹⁵N-labeled FliG_1–102 bound to unlabeled FliF_495–532 peptide. Red: TROSY-HSQC data of ¹⁵N-labeled FliF_495–532:FliG_1–102 fusion protein after TEV proteolysis. (B) Black: TROSY-HSQC data of ¹⁵N-labeled FliF_495–532:FliG_1–102 fusion protein before TEV proteolysis Red: TROSY-HSQC data of ¹⁵N-labeled FliF_495–532:FliG_1–102 fusion protein after TEV proteolysis. See also SI Figures 1, 2 and 6.

Solution-state structure of the FliF_C:FliG_N complex by SAXS

SAXS was used to evaluate the FliF_C:FliG_N solution-state structure. For these experiments, size-exclusion chromatography (SEC) was coupled to the SAXS measurements to separate different oligomeric or conformational states and limit contamination of soluble aggregates. The SEC chromatogram of the purified FliF_C:FliG_N complex (SI Figure 3A) reveals a well-behaved protein complex. SAXS was measured for the area enclosed by the black dotted lines on the SEC chromatogram shown in SI Figure 3A. Kratky analysis of the scattering data (SI Figure 3B) converges towards zero at high q values, characteristic of a globular protein complex (Lipfert et al., 2007). Generation of a low-resolution de novo molecular envelope of the protein complex results in a globular envelope with an extended region (SI Figure 3B). Interestingly, there appears to be poor agreement between the envelope and the structure of FliG_N from a full-length FliG structure determined from Aquifex aeolicus (SI Figure 3D; Lee et al., 2010). Either the binding of FliF_C to FliG_N produces a substantial conformational change in FliG_N or the FliF_C peptide is large enough to account for the unfilled area within the envelope. It is likely that FliF binding alters FliG, as the interaction converts FliG from a homodimer (FliG:FliG) to heterodimer (FliG:FliF) (Levenson et al., 2012). For more information regarding SAXS data acquisition and envelope generation, see SI Methods.

Crystal Structure of FliF_C:FliG_N complex

Crystals of the FliF_C:FliG_N complex were grown by vapor diffusion at pH 7.5 (see Methods). Attempts to determine the structure of FliF_C:FliG_N by molecular replacement (MR) failed, therefore a selenomethionine (seMet) variant of the FliF_C:FliG_N complex was crystallized (see Methods). The seMet FliF_C:FliG_N crystals diffracted to 2.6 Å resolution and yielded an initial SAD phased map with a figure of merit of 0.341. A de novo model was built and refined into the map with excellent final agreement statistics (Table 1). The unit cell consisted of two heterodimers per asymmetric unit, with one dimer having an extended α-helix towards the C-terminal portion of the FliG peptide, and the other having three segmented helices (Figure 3A). Although the two conformations of the last α-helix in FliG_N may indicate flexibility of this helix, the dimer with the extended α-helix (Figure 3) agrees better with the SAXS envelopes and is thus presumed to represent the dominant solution-state structure (SI Figure 4C).

Table 1.

Data collection and Refinement Statistics

	Wild Type FliFC:FliGN complex	Selenomethionine FliFC:FliGN complex
Wavelength (Å)	0.97700	0.97921
Synchrotron	CHESSa	APSb
Beam-line	A1	24-ID-E
Space group	P21	P21
a, b, c (Å)	49.18, 59.33, 51.72	48.81, 59.24, 51.76
α, β, γ (°)	90.00, 115.59, 90.00	90.00, 115.76, 90.00
Resolution (Å)	50.0 - 2.10 (2.15 - 2.10)	50.0 - 2.60 (2.66 - 2.60)
†Rmerge (%)	8.2	12.2
†Rp.i.m. (%)	5.5 (21.8)	5.3 (15.8)
†Rmeas (%)	9.9 (35.5)	13.3 (38.8)
▪I/σ(I)▪	13.4 (3.2)	14.8 (7.0)
Completeness (%)	99.5 (99.6)	99.2 (99.5)
Multiplicity	3.1 (2.4)	6.0 (6.0)
Anomalous completeness (%)		98.1
Mosaicity	0.43 – 0.81	0.63 – 2.55
Total Reflections	47,958	46, 599
Phaser FOM		0.341
Refinement
Resolution (Å)		44.35 - 2.10 (2.18–2.10)
No. of unique reflections		15,618 (1559)
Reflections used for Rfree		1556 (161)
Rwork/Rfree (%)		16.9 (17.4)/21.7 (23.7)
Clash Score		4.70
No. of non-hydrogen atoms		2263
Protein		2090 (260 residues)
Water		173
B factors (Å2)
Wilson		18.1
Average B-factors		24.3
Protein		23.7
Water		31.8
R.M.S. deviations
Bond lengths (Å)		0.007
Bond angles (°)		0.9
Ramachandran outliers (%)		0.4
Rotamer outliers (%)		1.3

Statistics for the highest-resolution shell are shown in parenthesis,

^†R_i = ΣΣ_i|I_i − <I_i>|/ΣΣ_i I_i

^aCornell High Energy Synchrotron Source - Cornell University

^bAdvanced Photon Source - Argonne National Lab

Figure 3.

Crystal Structure of FliF_C:FliG_N. The two unique heterodimers per asymmetric unit oriented as in the crystal (A) or with the same perspective (B) (FliF_{C -} violet and green, FliG_N - yellow and cyan). The two different morphologies of the terminal FliG_N helix are circled. See also SI Figures 3, 5, 6, and 7.

The FliF_C terminal peptide forms two α-helices connected by a short extended linker (F_Cα1 residues 497–514, and F_Cα2 517–529, Figure 3B). FliG_N is an all α-helical structure, similar to the all α-helical T. maritima FliG_M and FliG_C domains (Brown et al., 2002; Paul et al., 2011; Sircar et al., 2015), as well as the full-length FliG from A. aeolicus (Lee et al., 2010). FliG_N consists of four helices, denoted G_Nα1–4 in Figure 3B (residues 6–19, 21–30, 33–45, and 51–87, respectively) where G_Nα1–3 form an ARM-like domain that is also present in FliG_M and FliG_C (Brown et al., 2002; Lee et al., 2010). Together with F_Cα2, G_Nα1–3 comprise a 4-membered right-handed super-helix that has been well-documented as a structural motif in other FliG structures (Brown et al., 2002; Lee et al., 2010; Paul et al., 2011) and is also found in the N-terminal cytoplasmic domain of the Mg²⁺ transporter MgtE (Hattori et al., 2007). We will refer to these FliG super-helical domains as ARM-like motifs, following previous designations ((Lee et al., 2010; Vartanian et al., 2012), although we note that these domains differ substantially from traditional ARM motifs in helix crossing angles and packing interactions (Andrade et al., 2001). F_Cα1 and F_Cα2 assume a hook-like structure that latches into the V-shaped cavity formed from the G_Nα1–3 super helix and G_Nα4. In the unbound FliG_N of A. aeolicus (Lee et al., 2010), G_Nα1–3 forms a similar super helix but G_Nα4 is instead composed of two helices connected by a linker. Although similar in secondary structure, the unbound FliG_N domain G_Nα4 from A. aeolicus does not superimpose as well with FliG_N in complex with FliF_C (SI Figure 4A). In the absence of FliF_C this region may be highly flexible, which would explain the difficulty in crystallizing unbound FliG_N.

FliF_C and FliG_N associate to form a shared hydrophobic core. Interacting regions of FliF_C and FliG_N mapped by NMR involve the burial of conserved hydrophobic residues on both FliF_C and FliG_N (Levenson et al., 2012). Similarly, the electrostatic potential at the molecular surface of FliG_N reveals a hydrophobic, non-charged FliF binding region. The FliF_C:FliG_N complex buries ~1,700 Ų of hydrophobic binding surface on the FliF_C peptide, and is stabilized at the periphery by electrostatic interactions (SI Figure 5). A nearly invariant tryptophan residue essential for the FliF_C:FliG_N interaction (T. maritima FliF W527) (Thomas et al., 2001; Levenson et al., 2012) is tightly sequestered within a hydrophobic pocket formed from G_Nα1, G_Nα4, and F_Cα2 (SI Figure 5E).

From the extensive hydrophobic interactions it appears as though FliG_N folds around FliF_C. Surprisingly, the resulting topology matches that of the FliG_M and FliG_C domains (Figure 4A and 4B). The structure of the FliF_C:FliG_N complex superimposes well with the tertiary structure of FliG_M and FliG_C from T. maritima (Figure 4; left - FliG_M PDB: 3SOH, right - FliG_C PDB: 1LKV), producing RMS values of 1.053 Å (40 residues) and 1.227 Å (63 residues), respectively (Paul et al., 2011; Brown et al., 2002). Superposition of the FliF_C:FliG_N complex with the middle and C-terminal domains from A. aeolicus FliG_FL results in a similar outcome (SI Figure 4) (Lee et al., 2010).

Figure 4.

Domain repeats within FliG. (A) Superimposition of T. maritima FliF_C:FliG_N complex (FliF_C - violet, FliG_N - yellow) and Thermotoga maritima (left) FliG_M (grey, PDB 3SOH, RMS 1.053 Å, 40 residues) and (right) FliG_C (grey, PDB 1LKV, RMS 1.227 Å, 63 residues). The black circle signifies the absence of a corresponding ARM_C domain. (B) Comparison of secondary structure motifs and primary sequence alignment of the superimposed regions. See also SI Figure 4.

Mapping of the FliF_C:FliG_N interface by paramagnetic resonance enhancement

To further investigate the solution-state structure of the FliF_C:FliG_N complex, paramagnetic relaxation enhancement (PRE) experiments were carried out in which ¹H-¹⁵N TROSY-HSQC NMR spectra were collected on a FliF_495–532 (E508C):FliG_1–102 (E. coli FliF numbering) mutant that was conjugated to a MTSL nitroxide spin-label. Proximity to the conjugated MTSL causes neighboring ¹⁵N backbone nuclei to experience increased resonance peak broadening. Mapping of the maximally broadened residues in FliG_N from the spin label on F_Cα1 produces a binding interface that is in excellent agreement with the crystal structure (SI Figure 6 and SI Table 2). Notably, a small portion of the C-terminal tail of G_Nα4 also experiences a significant amount of resonance peak broadening with peak intensities 27–65% compared to that observed in the protein after the MTSL label had been reduced to a spectrally inactive form by addition of ascorbic acid, despite these residues being relatively far removed from the spin-label in the crystal structure. Perhaps the suspected flexibility of this helix brings it into proximity of the label for some conformational states of the complex.

Modeling of MS/C-Ring Interface of the Bacterial Flagellar Motor

The FliF_C:FliG_N complex was then docked into 3D cryo-EM reconstructions of the intact CW locked Salmonella motors (Thomas et al., 2006). It was assumed that the FliF_C peptide descends perpendicular to the inner membrane, despite the 32-residue gap present between F_Cα1 and the second transmembrane helix (TM2) region of FliF_NM. Symmetric rings were generated that contained either 24, 25, or 26 copies of FliF_C:FliG_N, consistent with previously determined FliF:FliG stoichiometry (Thomas et al., 2006). When FliF_C:FliG_N rings with 24 or 26 subunits were docked into the C₂₅ cryo-EM density, there were either clashes from overlapping secondary structural motifs (26 subunits), or unaccounted areas of electron density (24 subunits). Rings modeled with C₂₅ symmetry had minimal steric clashes within the ring, minimized areas of unassigned density and produced a correlation coefficient of 0.83 to 30 Å resolution (Figure 5A–C). Thus, the dimensions of the FliG_N:FliG_C complex agree well with the MS/C-ring symmetry. In modeling the MS/C-ring interface, F_Cα1 was directed upwards towards the inner-membrane, an orientation that conferred biological relevance, as well as allowed for systematic packing of all 25 subunits within minimal steric clashes (Figure 5A–C). In agreement with the model, recent bioinformatics-based modeling of the upper periplasmic region of FliF_NM indicates 25 FliF subunits per MS-ring (Bergeron, 2016).

Figure 5.

Modeling of the FliF_C:FliG_N ring with C₂₅ symmetry. (A) Fully assembled (FliF_C:FliG_N)_n ring (n = 25 copies) showing the side (left) and top (right) view. (B) (FliF_C:FiG_N)₂₅ ring and (FliM_M:FliG_MC)₃₄ ring, (red -FliM_M, blue -FliG_MC) superimposed with the EM density of the Salmonella rotor (Thomas et al., 2006) Ring density simulated at a 30 Å from the structures gave correlation coefficients with the EM density as follows: (FliF_C:FliGN) 0.83, and (FliM_M:FliG_MC) 0.79.

In an effort to extend the modeling further, we carried out the same fitting technique with a FliG_MC:FliM_M complex structure previously generated via x-ray crystallography and pulse-dipolar ESR (PDB ID 4QRM, Sircar et al., 2015). In total, 34 copies of a single FliG_MC:FliM_M complex were fit into the rotor density, producing a correlation coefficient of 0.79 (Figure 5C). Notably there remains a disparity between the FliG_MC and FliG_N stoichiometry by 9 copies. This symmetry mismatch is well recognized (Thomas et al., 1999, 2001; Manson, 2007; Berg 2003; Sowa et al., 2008), and could result from unoccupied positions of FliG_MC in the C-ring (Sircar et al., 2015), non-equivalence in the FliG:FliM contacts (Paul et al., 2011; Sakar et al., 2011) or as a result of adaptive remodeling of both FliM and FliN (Fukuoka et al., 1994; Lele et al., 2012; Delalez et al., 2014). Nonetheless, the current model suggests that 34 copies of FiG_N could not fit within the upper portion of the EM density.

Functional analysis of the FliF_C:FliG_N complex

To validate the structure of FliF_C:FliG_N, site-directed mutants were evaluated by in vivo motility assays. All experiments were performed in E. coli with homologous E. coli proteins, where FliF and FliG share 29% and 36% sequence similarity with the T. maritima proteins, respectively, and conserve nearly all of the key residues involved in the complex interface (SI Figure 7) (Levenson et al., 2012; Thomas et al., 2001). Additionally, residue substitutions at the FliF-FliG interface were tested in protein interaction studies and crosslinking of engineered cysteine mutants were further employed to probe the functional FliF:FliG interface.

Hydrophobic to aspartate mutations FliG were made and cell migration rates were measured relative to wild type (Figure 6A,B). Residue substitutions along the FliF:FliG interface observed in the crystal structure proved most detrimental to motility. Specifically, motility defects were strongest for changes that are in close proximity to F_C α 2, especially at positions Leu13, Leu29, Leu37, Ile17 (E. coli numbering), which is consistent with the interactions between FliF_C and FliG found in the crystal structure.

Figure 6.

In vivo E. coli motility and interaction assays. (A) Two views of the FliG_N domain, showing motility defects that result upon replacement of the indicated hydrophobic residues with aspartate. Motility phenotypes were tested by expressing mutant FliG proteins in the fliG null strain. Yellow-green, rate less than 100% but greater than 70% of wild-type; yellow, rate between 20% and 70% of wild type; orange, motile but at less than 20% of wild type; red, immotile. (B) A table summarizing the relative rates, values are the mean ± s.d. for 3 determinations. (C) Interaction between FliF_C (residues 505–552) and FliG_N (residues 1–87) observed in the bacterial adenylate cyclase two-hybrid system. Positive interactions drive expression of β-galactosidase. Negative controls and a leucine-zipper positive control are shown. (D) Effects of hydrophobic-to-aspartate replacements in FliG_N on the FliF-FliG interaction. Coloring is based on β-galactosidase activities: Green, activity indistinguishable from wild type; yellow-green, activity less than 100% but greater than 70% of wild type; yellow, activity between 20% and 70% of wild type; red, activity decreased to the level of negative controls. FliF_C shown light purple in all figures. See also SI Figure 8.

Analysis with the bacterial adenylate cyclase two-hybrid (BACTH) method provided independent support for a strong FliF:FliG interaction involving the interfaces observed in the crystal structure (Figure 6D) (Miller et al., 1972; Karimova et al., 2001; Battesti et al., 2012). BACTH experiments used hybrid constructs containing the FliF_C (E. coli residues 505–552) and FliG_N (E. coli residues 1–87). In experiments with wild-type FliF_C and FliG_N, strong color development was observed on MacConkey plates, and measurements of β-galactosidase activity afforded values comparable to the leucine-zipper positive control (Figure 6C). Hydrophobic-to-aspartate replacements of FliG on the predicted interface weakened the interaction in a pattern similar to the motility phenotypes. Residue substitutions made to FliF provided similar validation of the expected contact sites (SI Figure 8). These results suggest that the interface involving F_Cα2 may be chiefly responsible for the strength of the interaction. Consistent with this, a hydrophobic-to-aspartate replacement at FliF residue 542 in F_Cα2 eliminated both motility and the two-hybrid interaction. A mutation at position Val538 in F_Cα1 caused a significant reduction in migration rate, while not diminishing the FliF:FliG interaction. In liquid culture, this mutant exhibited relatively weak motility and a substantial fraction of immotile cells, indicating a delay or partial defect in flagellar assembly. Thus, the V538D replacement alters the FliF:FliG relationship while not greatly weakening the interaction.

To further probe the FliF:FliG interaction in vivo, cross-linking of engineered cysteine mutants were employed to map the binding site interface between FliF and FliG in E. coli. Targeted disulfide crosslinking experiments indicate that the FliF:FliG relationship in E. coli is similar to that observed in the structure of the T. maritima proteins (Figure 7C). Fifty-five double-cysteine mutants were made (Figure 7A; SI Table 3). In total, seven relatively strong crosslinks were observed, all involving positions that are in proximity in the crystal structure (C_β-C_β distances 6–10.5 Å). All of the cysteine double mutants that displayed strong crosslinking retained significant function (45% – 85% of wild-type) in soft-agar. Crosslinking experiments yielded one surprising result; Cys543 in FliF showed only weak crosslinking though predicted to be close to several of the FliG_N cysteine replacements. The strongest crosslink of Cys543 (still weak in absolute terms) was to Cys29 in FliG, which according to the structure is the closest (to 543) of the 11 FliG replacements examined. Residue 543, normally Ile, resides in F_Cα1 and is buried at the FliG:FliF interface. We hypothesized that packing in this region might be so stable as to prevent deprotonation or necessary movements of the buried cysteine side chains for crosslinking. In an attempt to loosen the interface and increase the reactivity of Cys543, the adjacent Arg544 was replaced with alanine; however, crosslinking was not significantly increased. Crosslinking was then carried out in the presence of urea (Figure 7B). The crosslink to FliG at Cys29 was slightly enhanced in the presence of 3 M urea, while crosslinking to a more distant position (Cys44) remained negligible.

Figure 7.

In vivo FliG-FliF crosslinking. (A) Products of iodine-induced disulfide crosslinking were detected using anti-FliG immunoblots. The seven cysteine pairs that gave strong crosslinking are shown together with single cysteine controls. Crosslinking was weak or undetectable for forty-eight other cysteine pairs tested (SI Table 3). In addition to the indicated cys-mutant FliF, a comparable amount of wild-type FliF (expressed from the chromosome) was present in the crosslinking experiments; thus, crosslinking yields should underestimate what would occur if all of the FliF carried the Cys replacement. (B) A low-yield crosslink between FliG position 543 and FliG position 29 that is made somewhat stronger (by ~50%) in the presence of 3 M urea. The more-distant position Cys44 did not crosslink. (C) Mapping the crosslinks onto the FliG-FliF structure. Thicker red lines indicate the crosslinks observed under standard conditions, and the thin red line indicates the weaker, urea-enhanced crosslink involving the buried position 543. See also SI Figure 8.

Overall, results from the in vivo experiments fit well with the crystallographic structure of the FliF_C:FliG_N complex, and indicate that the FliF:FliG interfaces are essentially identical in E. coli and T. maritima. As expected, the FliF:FliG interaction is critical for function. Mutant phenotypes and the non-reactivity of Cys543 highlight the importance of the interactions involving F_Cα2 and indicate that packing at this interface may be unusually stable.

Discussion

The MS/C-ring interface is an essential contact site of the flagellar motor. Not only does it serve as a checkpoint during flagellar morphogenesis, but it positions FliG with respect to the inner membrane such that it can (1) bind to the MS-ring to propagate rotation, (2) interact with the stator complexes to confer torque generation to the switch complex, and (3) position FliG such that it can bind to FliM to anchor CW/CCW switching and rotor dynamics (Berg, 2003; Sowa, 2008). Although fusion-deletion and mutagenesis studies have established that FliF and FliG interact through their N and C terminal domains, respectively, a molecular level view of this interaction remained elusive. The 2.1 Å resolution crystal structure of FliF_C:FliG_N defines the interface between the MS and C-rings within the flagellar motor. Furthermore, by reconciling the FliF_C:FliG_N crystal structure with NMR, SAXS, in vivo assays and modeling to cryo-EM reconstructions, we validate the structure as biologically relevant and further expand current understanding of protein organization at the MS/C-ring interface.

Fusion of the FliF_C peptide to FliG_N was an effective strategy for producing the complex. Comparing TROSY-HSQC ¹H-¹⁵N NMR spectroscopy of the fused protein to the complex formed by its components indicates that the fusion folds as the native complex (Figure 2). Given the intimate contact formed by the two separate proteins, it is not surprising that they appear to fold as one unit. Notably, the ability of the cleaved and non-cleaved moieties to form essentially the same structure places spatial restrictions on the location of the FliF_C binding site. Fortunately, these restrictions were satisfied in the native FliF_C:FliG_N complex (Figure 3), and found to be in agreement with the FliF_C:FliG_N interface mapped from the PRE studies (SI Figure 6 and SI Table 2).

FliF_C:FliG_N crystallizes in two conformations (Figure 3A). It was important to consider both when modeling the MS/C-ring interface. The FliG_N core that houses the FliF_C binding site remains invariant, owing to the strength of the FliF:FliG interaction. However, the C-terminal tail of FliG_N (G_Nα4) assumes either an extended α-helix, or three segmented helices (Figure 3B). Although the crystal lattice may influence these conformations, the solution-state complex favors the extended helix (SI Figure 3B–C). Nonetheless, the more compact conformation may play a role in motor operation, as it confers a higher degree of flexibility thus could serve as a flexible hinge to absorb stress during rotation reversals. Modeling of the FliF_C:FliG_N structure into the ring density (Figure 5A–B) directs F_Cα1 toward the inner membrane of TM2 of FliF. The 32 unaccounted residues between FliF_C F_Cα2 and TM2 of FliF_NM could similarly serve as a flexible hinge.

Importantly, FliG_N requires FliF_C to fold into a stable FliG_M/FliG_C-like domain, and thus the stoichiometry of FliF to FliG is most likely to be 1:1. Such a 1:1 complex has important implications for the overall structure of the rotor, as FliG must interact with FliM, of which there are ~34 copies. Therefore, either not every FliM interacts with one FliG in the same way, or the protein stoichiometry in the MS and C-rings are not correct. The latter seems unlikely, as 25 copies of the FliF:FliG complex fit well around the upper switch, and it is difficult to accommodate more subunits without significant steric clashes. Further, the current finding that the MS/C-ring interface has C₂₅ symmetry is in agreement with recent homology modeling of the periplasmic region of FliF (Bergeron, 2016).

FliG is divided into three distinct domains, where each domain has evolved to carry out a specific function within the flagellar switch (Lloyd et al., 1996; Berg, 2003; Sowa, 2008). FliG_N contains the binding site for FliF_C and anchors the C-ring to the inner-membrane, FliG_M interacts with FliG_C and FliM_M to propagate CCW/CW switching, and FliG_C contains conserved charged residues to interact with membrane-bound stators in torque generation (Lloyd et al., 1997; Berg, 2003; Sowa, 2008). The structure of FliG determined from A. aeolicus agrees with structures of the M and C terminal domains (Brown et al, 2002; Paul et al., 2002). A notable feature of unbound FliG_N identified in the full-length structure was the structural similarity between AqFliG_N helices N1–4, and AqFliG_C terminal helices 3–6 (Lee et al., 2010). A similar observation was made previously from the crystal structure of FliG_MC from T. maritima (PDB 1LKV) regarding the ARM motifs present in FliG_M and FliG_C (ARM_M/C) (Brown et al., 2002). Interestingly, the FliF_C:FliG_N complex shows how the structural replication within FliG_C is even more extensive than in FliG_MC (Figure 4; SI Figure 4). When FliF is bound to FliG, all three domains adopt similar conformations such that FliF:FliG aligns well with the ARM_M motif and helix_MC/NM in FliG_M, as well as with helices 1–6 in FliG_C. The striking similarity of the FliF:FliG fold with that of FliG_M and FliG_C suggests domain shuffling may relate all three of the FliG domains (Di Roberto et al., 2014). Either the N-terminal helix of FliG_N was transferred to the C-terminus of FliF, or the FliG_N:FliF_C unit was fused and propagated to generate FliG_M and FliG_C. The genes for FliF and FliG are often adjacent, in the same orientation and expressed from the same promoter. Thus, in many instances the coding sequences for the two helices that complete the FliG_N superhelix are tightly coupled to the FliG gene. Furthermore, in Chlamydia the C-terminal region of FliF has been replaced with an entire FliG_M-like domain (Bergeron, 2016). Thus, the interface between the FliF TM2 and the C-ring is always a FliG_M-like domain, whether it is composed of a single polypeptide or two that mate FliF_C to FliG_N. Interestingly, upon comparing the sequences of the FliF_C:FliG_N complex to FliG_M, the region where FliF_C terminates and FliG_N begins coincides with the FliM_M-binding EHPQR motif of FliG_M (Paul et al., 2011; Vartanian et al., 2012) (Figure 4B). If FliG_M and FliF_C:FliG_N share a common origin, severing a continuous FliG_N domain into two at what would be the FliM interaction site on the analogous FliG_M may have been advantageous to prevent incorrect binding of FliG_N to FliM_M during MS/C-ring assembly.

The contact residues of the FliF:FliG structure mediate protein interactions in cells and are critical for generating functional flagella. Substitution of key hydrophobic residues to aspartic acid along the FliF:FliG interface of both FliG (Figure 6A) and FliF (SI Figure 8) disrupt cell motility and protein-interactions. Although F_Cα1 buries ~570 Ų of surface area, the functional specificity and strength of the FliF:FliG interaction can largely be attributed to F_Cα2. Disulfide crosslinking data provides direct biochemical evidence in support of these findings. One surprising observation was the relatively small extent of crosslink formation between buried FliF Cys543 and any potential disulfide partner (Figure 7B, SI Table 4). This result likely reflects the uniquely strong interaction between F_Cα2 and FliG that prevents movement of the cysteine residues or access to oxidants. These results provide evidence for the conjecture that FliF_C and FliG_N essentially fold as one domain to provide a robust anchor for the C-ring to the membrane. It is not surprising then that the FliF:FliG interaction is much stronger than that of the FliG:FliM or FliM:FliN and can persist longer at low pH without dissociation (Francis et al, 1994).

Overall, by combining data from x-ray crystallography, SAXS, NMR, modeling to cryo-EM structures and in vivo functional studies, we have defined the FliF:FliG interaction at the MS/C-ring interface. The structure has then been elaborated into an updated model for the upper switch. Importantly, the structure suggests that the FliF:FliG stoichiometry is likely the same, and that domain shuffling appears to have driven the association of these key building blocks of the flagellar motor. We propose a model where a domain is duplicated and then genetically split so that a folding unit can associate two otherwise non-interacting proteins with high affinity. It is notable that protein engineering has used a similar strategy of split protein complementation to effect interactions in cells through the development of tools such as split GFP, ubiquitin and luciferase proteins. (Xing et al., 2016)

Experimental Procedures

i. Protein Expression and Purification

Native proteins were expressed in BL21 (DE3) E. coli cells, grown at 37°C in LB media containing 50 μg-mL⁻¹ kanamycin, and induced with 1 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG) at 25°C after reaching an OD₆₀₀ of ~0.6. Selenomethionine variants were expressed in B834 methionine auxotrophic E. coli cells at 37°C in minimal media containing 100 μg-mL⁻¹ ampicillin hydrochloride and 50 mg-L⁻¹ +/− L-selenomethionine.

All proteins were purified according to the following procedure: cell pellets were re-suspended in lysis buffer (25 mM HEPES pH 7.5, 500 mM NaCl, 5 mM imidazole), lysed via sonication, and the lysate was then centrifuged at 20,000 rpm for 45 minutes at 4°C. The supernatant was passed through a Ni-NTA column, washed with 50 mL wash buffer (25 mM HEPES pH 7.5, 500 mM NaCl, 25 mM imidazole), and eluted with elution buffer (25 mM HEPES pH 7.5, 500 mM NaCl, 200 mM imidazole). The linker between FliG_N and FliF_C was removed by overnight incubation with TEV protease at 30°C. TEV was removed via centrifugation and the FliF_C:FliG_N complex was dialyzed extensively against 50 mM sodium phosphate pH 6.5. For the preparation of selenomethionine FliF_C:FliG_N, all buffers were supplemented with 10 mM DTT. Successful selenomethionine incorporation was confirmed via whole-protein positive ion electrospray LC-MS/MS.

Crystallization, Data Collection, and Phasing

The FliF_C:FliG_N complex was subjected to sparse course matrix screening in 600 nL drops at a protein concentration of 65 mg/mL. Crystals grew in 100 mM HEPES pH 7.5, 25% (w/v) PEG 3000, 200 mM NaCl at 25°C after ~7 days. Selenomethionine-incorporated crystals grew from 33 mg/mL protein, 100 mM imidazole pH 7.2, 130 mM NaCl, 30% (w/v) PEG 8000 in ~14 days.

Briefly, data sets were integrated and scaled in HKL2000 and SCALPACK (See Table 1). Phasing of the anomalous data set was done with Phenix HYSS and the model built with Autobuild (Phenix) programs (Adams et al., 2010). Manual adjustments to model were made with COOT and subsequent refinements were carried out with Phenix Refine. The final FliF_C:FliG_N model has an R_free of 21.74 % and a R_work of 16.92 % with excellent stereochemistry.

ii. Ring Simulation and Fitting into cryo-EM Maps

Low-resolution cryo-EM maps of whole rotor density containing the C, MS, L, and P rings were provided by Dennis Thomas (Thomas et al., 2006). To construct a model of the upper C-ring/MS-ring interface, a single FliF_C:FliG_N complex was aligned to the whole rotor density such that the coordinate systems of the crystal structure PDB file coincided with that of the rotor MRC map file. The center of mass was calculated for the placed complex, and the structure was duplicated n number of times and rotated 360/n degrees around an axis chosen perpendicular to the membrane. The ring was simulated to a radius of 15 nm, a distance corresponding to the radius of the EM map. Three rings of n = 24, 25 and 26 C_n fold symmetry were generated and evaluated with respect to the EM density. By inspection, only rings of C₂₅ fold symmetry produced a clash-free model that has minimal gaps in density when comparing. The C₂₅ upper C/MS-ring model was fit into the EM map using the Fit to Map program in Chimera. When simulating the EM map to a resolution of 30 Å, the C₂₅ fold FliF_C:FliG_N ring provided good agreement with the whole rotor density and produced a correlation coefficient of 0.83. The above procedure was repeated with a FliG_MC:FliM_M model (PDB 4QRM), to give a mid rotor subunit copy number of 34 and a correlation coefficient of 0.79.

iii. Generation, purification, and NMR analysis of isotopically enriched FliF_C:FliG_N complex

The fusion FliF_C(495–532):FliG_N(1–102) construct was generated by traditional molecular cloning methods. The FliF_C portion was amplified from a plasmid obtained from the JCSG and cloned into a plasmid containing FliG_N(1–102) (Levenson et al., 2012). All NMR data was collected with a construct similar to SI Figure 1, however, the construct only included one TEV protease site between the His₈ tag and the FliG_N domain. Isotopically-labeled proteins were grown in M9 minimal media supplemented with ¹⁵NH₄Cl, ¹³C glucose, and/or D₂O (when necessary). Rosetta (DE3) E. coli cells were used for expression of all constructs. Cells were grown at 37 °C, and upon reaching an OD₆₀₀ of ~0.6, were induced with a final concentration of 1 mM IPTG and expressed for approximately 6 hours. Purification was identical as described previously. After SEC, the protein samples were dialyzed into 50 mM sodium phosphate, 100 mM NaCl, pH 6.5 for NMR analysis. After dialysis, all samples were concentrated to their final concentration, at which point 0.02% (w/v) NaN₃ was added. NMR spectra were collected on a Bruker Avance 800 MHz spectrometer equipped with a cryoprobe.

E. coli strains and media for interaction and motility studies

Escherichia coli strains and plasmids are listed in SI Table 1. All strains were derivatives of the wild type strain RP437. Chromosomal in-frame deletions or point mutations were made by using the lambda red method (Datsenko et al, 2000). TB medium contained (per liter) 10 g tryptone and 5 g NaCl, LB medium contained the same plus 5 g yeast extract. Soft-agar motility plates used TB media and Bacto-agar at 2.6 g-liter⁻¹. Ampicillin was used at 100 μg-ml in liquid media, 100 μg-ml⁻¹ in selective plates, and 50 μg-/ml⁻¹ in soft-agar motility plates. Chloramphenicol was used at 50 μg-ml⁻¹ in liquid media and selective plates and at 25 μg-ml⁻¹ in motility plates. IPTG and sodium salicylate were prepared as aqueous 0.1 M and 10 mM stocks and used at the concentrations indicated in the figures.

iv. Motility assay

All of the experiments examining motility defects were carried out in the respective null backgrounds (DB225, EKS10, see SI Table 1). Soft-agar plates were spotted with 3 μl of overnight cultures. Once migration began, colony size was measured at regular intervals and plots of diameter vs. time fitted to a line to determine rates. Rates are reported relative to wild-type controls measured in the same experiment. Effects of Asp replacements on motility were measured in strains expressing the mutant FliG from a plasmid (derivatives of pKP619), induced with 100 μM IPTG. Effects of Asp replacements in FliF were measured in strains expressing FliF from the plasmid pEK16, induced with 0.3μM salicylate, and additionally expressing wild-type FliG from plasmid pKP619 (induced as above), in a ΔfliF strain. The additional, plasmid-expressed FliG was found to be necessary for optimal function. Effects on motility of cysteine replacements in FliG and FliF were measured in a ΔfliFfliG double-deletion strain, with both proteins expressed from plasmids.

v. In vivo and crosslinking mutagenesis

Site-directed mutations were made using the QuikChange method (Stratagene). DNA sequencing and oligonucleotide synthesis were carried out by core facilities at the University of Utah.

vi. Double-cysteine cross-linking

Disulfide crosslinking between FliF and FliG was studied in strains expressing cysteine-containing FliG variants from the chromosome and cysteine-containing FliF variants from plasmid pEK16. For these experiments, wild type protein (FliF) was present along with the Cys-mutant FliF. Levels of the two were roughly equal. The plasmid was induced with 0.2 μM salicylate, which gave FliF levels sufficient to compete with the wild-type (cysteine-less) FliF expressed from the chromosome but not so high as to impair motility. (Motility impairment occurred with induction by 1 μM or higher.) Cells were cultured overnight at 37° C and diluted 100-fold into TB containing antibiotic and 0.2 μM sodium salicylate then grown at 32° C for 5–6 h (to OD₆₀₀ 0.7 – 0.8). OD₆₀₀ was measured to adjust the cell density, then equal numbers of cells were pelleted and re-suspended with XL buffer (20mM Na-phosphate pH7.4, 150mM NaCl). For cross linking with iodine, 0.1ml of cell mixed with 4 μl of 25 mM iodine for 3 min at room temperature, then sulfhydryl groups were blocked by addition of 2 μl 0.5 M NEM and incubation at room temperature for 3 min. Samples were mixed with an equal volume of 2X non-reducing loading buffer and heated at 95 °C for 10 min before loading on SDS-PAGE gels. Some experiments included urea during the cross-linking step, at the concentrations indicated in the figures, to test the effects of partially destabilizing the proteins.

vii. SDS page and immunoblotting

Proteins were resolved in 7.5% SDS-PAGE gels and transferred onto nitrocellulose using a Transblot turbo apparatus (BioRad). Rabbit polyclonal antibody against FliG was used at 1:1,000 dilution in a solution containing phosphate-buffered saline pH 7.4, 0.1% gelatin, and 0.01% sodium azide. Immunoblots were visualized and analyzed using the LiCor Odyssey infrared-imaging system.

viii. Two-hybrid interaction assay

Bacterial adenylate cyclase two-hybrid (BACTH) measurements of the FliF:FliG interaction used vectors provided in a kit from Euromedex. Procedures followed previously published procedures (Miller et al., 1972; Battesti et al., 2012) with minor modifications. Cells from a single colony were grown overnight at 32° C in LB containing ampicillin and 0.5 mM IPTG, then pelleted and re-suspended in PBS buffer. 0.04 ml of cells were mixed with 0.96 ml Z buffer (0.06 M Na₂HPO₄, 0.04 M NaH₂PO₄, 0.01 M KCl, 0.001 M MgSO₄, pH 7.0, 0.08% SDS, 0.22% (v/v) BME). Cells were permeabilized by addition 100 μl chloroform and vortexed for 10 sec, followed by incubation at 30° C for 5 min. Reaction was started by adding 0.2 ml ONPG solution (4mg-ml⁻¹). After measured times of reaction at 30° C, reaction was stopped by addition of 0.5 ml 1 M NaHCO₃. Product was quantified by measurement of OD₄₂₀, with a correction due to cell scattering. Activity in Miller Units was computed according to the formula: {1000 × [(OD₄₂₀ − 1.75 × OD₅₅₀)]/(time of reaction × volume of culture used × OD₆₀₀)}; where OD₄₂₀ and OD₅₅₀ are read from the reaction mixture and OD₆₀₀ reflects cell density in the washed cell suspension.

Highlights.

• FliF_C:FliG_N fold together to produce a topology that repeats throughout FliG

• FliF_C:FliG_N interact through hydrophobic contacts critical for motor function

• FliF:FliG 1:1 stoichiometry produces an MS/C-ring interface with ~C₂₅-fold symmetry