Subunit rotation is the mechanochemical intermediate for the catalytic activity of the membrane enzyme FoF1-ATP synthase. smFRET (single-molecule FRET) studies have provided insights into the step sizes of the F1 and Fo motors, internal transient elastic energy storage and controls of the motors. To develop and interpret smFRET experiments, atomic structural information is required. The recent F1 structure of the Escherichia coli enzyme with the ϵ-subunit in an inhibitory conformation initiated a study for real-time monitoring of the conformational changes of ϵ. The present mini-review summarizes smFRET rotation experiments and previews new smFRET data on the conformational changes of the CTD (C-terminal domain) of ϵ in the E. coli enzyme.
Introducing the rotary motors of FoF1-ATP synthase
The rotary engine FoF1-ATP synthase is the molecular protein machine  making most of the ATP in living cells. The ubiquitous multi-subunit enzyme is located in the plasma membrane of bacteria, the thylakoid membrane in photosynthetic cells and the inner mitochondrial membrane of eukaryotes. The enzyme operates as a mechanochemical energy transducer comprising two motors with different step sizes . The current assignment of rotor and stator subunits is shown in Figure 1(A). The F1 part of the enzyme catalyses the reaction of ADP plus Pi to ATP (ATP synthesis) and the reverse reaction (ATP hydrolysis) at three nucleotide-binding sites, and comprises the stator subunits α3β3δ and the rotary subunits γ and ϵ (Escherichia coli nomenclature is used for subunit names and residue numbers). The membrane-embedded Fo part translocates protons (or Na+ in some organisms ) associated with a rotation of the ring of c-subunits (ten in E. coli) with respect to the stator complex of a- and b2-subunits. According to this model, a full rotation of the proton-driven c-ring in Fo is subdivided into ten steps, but the attached γϵ rotor of F1 induces three sequential open-and-close movements of the nucleotide-binding sites in a three-fold symmetry of α3β3, i.e. in 120° steps. The intrinsic mismatch in symmetry and step angles is accommodated by transient elastic deformations  and reversible twisting of rotor subunits . The stator connection between the F1 and Fo motors (the b2δ subunits of E. coli FoF1), seen in electron micrographs as a peripheral stalk [5,6], is much more stiff, as determined from X-ray crystallography [7,8] and bead-rotation assays . In bacterial enzymes this could be due to the unusual right-handed coiled-coil structure of the b2 dimer .
E. coli FoF1-ATP synthase architecture, and cysteine positions for smFRET to monitor rotary subunit movements and ϵ conformational changes
Subunit rotation within the enzyme was predicted by P. Boyer about 30 years ago, based on subunit asymmetry and the co-operative behaviour of alternating catalytic sites (reviewed in ). Since then, structural studies (and biophysical methods) have supported subunit rotation, beginning with the ‘mother of all F1 structures’ published by Walker and colleagues in 1994 . Many subsequent mitochondrial F1 structures revealed atomic details of the catalytic process in the nucleotide-binding pocket and further supported the motor view of γ-subunit rotation.
The mode of c-ring rotation in Fo was inferred [10–13] from structural information using chemical cross-link data between introduced cysteine residues in the a- and c-subunits and NMR structures of isolated c-subunits . Recently, after successful crystallization of c-rings from different organisms, consisting of 8–15 subunits [15,16], atomic simulations of conformational dynamics supported the proposed essential elements of the Fo motor, i.e. electrostatic forces at the interface of the a-subunit and adjacent c-ring and a rotational swivelling motion of the proton-binding and -releasing transmembrane helix of the c-subunit (reviewed in ).
Biochemical evidence for subunit rotation was first provided by using hybrid F1 complexes and reversible intersubunit cross-linking to show different orientations of the F1 γ rotor with respect to the stator during catalysis in vitro [17,18]. An advantage to the approach was that it could be applied to membrane-embedded FoF1 to demonstrate changes in rotor orientation during ATP synthesis or hydrolysis. This was later applied to demonstrate that subunit ϵ also moves as part of the rotor . A similar cross-linking approach provided the first evidence for energy-driven rotation between the c-ring and the a-subunit of FoF1 in E. coli membranes . The disadvantages of these approaches were that they could not measure rotation kinetics or directionality.
The real-time kinetics of γ-subunit rotation were assessed in a spectroscopic experiment . Photoselection by polarized excitation was used for reversible photobleaching of a subset of surface-immobilized F1 parts, and γ-orientation-dependent fluorescence of covalently attached eosin molecules served as the marker of rotation. ATPase-driven changes revealed the rotary movement in milliseconds. However, the direct demonstration of γ-subunit rotation by video microscopy in 1997  paved the way for high-resolution biophysical measurements of single F1 motors (reviewed in ). The movement of the attached micrometre-long actin filament magnified the nanometre changes for light microscopy with its diffraction-limited resolution of approximately 200 nm.
To monitor γ-rotation, the α3β3γ subcomplex was prepared separately and immobilized on a glass surface. Therefore this approach cannot be used to analyse subunit rotation during ATP synthesis which is driven by PMF (protonmotive force) across the lipid bilayer. Very small markers are needed to observe rotation in FoF1-ATP synthase in the physiological membrane environment of living cells. Owing to the inherent structural asymmetry caused by the peripheral stalk of FoF1, synchronizing rotor subunit orientations is impossible in vivo. The promising biophysical method for obtaining information about ATP synthesis in vitro and in vivo is the real-time measurement of distance changes within a single enzyme, which requires two different small fluorophore molecules to be attached specifically to one rotor and one stator subunit. During movement of the rotor, the fluorophore distances can be followed in single enzymes based on FRET . Results of analysing time trajectories of subunit rotation by smFRET (single-molecule FRET), which are complementary to structural snapshots, are summarized here. The present mini-review of our current understanding of the motors and controls of single E. coli FoF1-ATP synthase ends with a brief preview of new smFRET evidence for the mechanism of blocking functional rotation by the ϵ CTD (C-terminal domain) (see conformations in Figures 1B and 1C).
smFRET for subunit rotation in FoF1-ATP synthase
The use of smFRET to measure conformational changes in proteins and nucleic acid complexes has become an increasingly popular and powerful microscopy method since its first proof-of-principle demonstration by Ha et al. published in 1996 . With smFRET one can measure fluorophore distances between 2 and 8 nm precisely with 1 Å (1 Å=0.1 nm) resolution (but broadened to approximately 5 Å resolution by stochastic movements of the FRET fluorophores along their linkers ) and with sub-millisecond time resolution . We were interested in time trajectories of subunit rotation in single liposome-reconstituted FoF1-ATP synthase. These proteoliposomes allowed the creation of a PMF for ATP synthesis conditions using the established buffer mixing approach of the Gräber laboratory . For the first successful smFRET rotation experiment with FoF1-ATP synthase shown in 2001 , the FRET donor fluorophore TMR (tetramethylrhodamine) was placed on the rotating γ-subunit to an introduced cysteine residue, which was considered to be located far away from the axis of rotation. The FRET acceptor fluorophore Cy5 (indodicarbocyanine) was attached to one of the peripheral b2 subunits. In the presence of 1 mM ATP and Mg2+, subunit rotation was inferred from stepwise FRET-distance changes in sequential order for a single FoF1-ATP synthase in the laser focus . For subsequent smFRET of the F1 and the Fo motor, different positions on the rotor subunits with respect to distinct positions on stator subunits were used [31–35].
Figure 2 summarizes the actual confocal smFRET measurement and analysis methods using freely diffusing proteoliposomes in buffer solution. Two laser foci are aligned to the same location for alternating excitation of the FRET fluorophores and, as an independent control , for the FRET acceptor only. When a FRET-labelled enzyme in a liposome traverses these excitation volumes due to Brownian motion, FRET donor excitation results in a burst of photons from FRET donor and acceptor (‘blue laser focus’ in Figure 2A). Nanoseconds later, the FRET acceptor is excited by the second laser (‘red laser focus’) to test whether this fluorophore is bound to the same enzyme and in order to exclude photophysical artefacts such as spectral fluctuations of the FRET donor fluorophore. For each data point in the photon burst, the fluorophore distance rDA can be calculated from the FRET efficiency according to EFRET=IA/(IA+ID)=R06/(R06+rDA6), using ID and IA, intensities of FRET donor and acceptor fluorophores (corrected for background, spectral cross-talk to the other detection channel, detection efficiencies and fluorescence quantum yields), and R0, Förster radius of the given fluorophore pair for 50% energy transfer. Within a photon burst, stepwise changes in EFRET indicating conformational changes or rotary movements of a subunit respectively have to be found either be manual inspection  or computationally, for example by hidden Markov models [38–40] or change point algorithms . Then, the following information about the motors of FoF1-ATP synthase is obtained.
smFRET of ϵ rotation in FoF1-ATP synthase
Opposite direction of motor rotation during ATP synthesis and hydrolysis
Stepwise changes of FRET efficiencies have been observed for smFRET measurements between the rotary ϵ-subunit of F1 and the stator b2 of Fo [35,42] (shown in Figures 2B and 2C). Three FRET levels called ‘L’ (low EFRET), ‘M’ (medium EFRET) and ‘H’ (high EFRET) with a sequential order of →H→M→L→H→ during ATP hydrolysis, but in reverse order →H→L→M→H→ for ATP synthesis, indicated the opposite direction of rotation for the distinct catalytic processes, as reported first for γ-subunit rotation in FoF1-ATP synthase in 2004 . Each FRET level was consistent and transitions occurred within approximately 200 μs .
Different rotary stopping angles during catalysis
Given the geometrical constraints for the rotary motion of ϵ or γ in F1, i.e. a 120° stepping at high (ATP) for ATP hydrolysis or high PMF for ATP synthesis, the three stopping positions of the rotary subunits with respect to b2 were very similar for the two catalytic modes as well as in the presence of AMP-PNP (adenosine 5′-[β,γ-imido]triphosphate)  (Figure 2D). However, in the presence of ADP and Pi, but without PMF, three distinct stopping positions L*, M* and H* were found (Figure 2E). This correlated with a cryo-EM study of E. coli F1 with a nanogold label on the ϵ N-terminal domain: only with ADP and Pi present, ϵ showed a distinct position relative to α- and β-subunits . The recently determined crystal structure of ϵ-inhibited E. coli F1  is also consistent with the distinct stopping positions of ϵ seen by smFRET. That is, whereas the main rotary pause should be at the catalytic dwell angle during catalytic turnover with excess substrate, ϵ-inhibited F1 appears to be paused at a position corresponding to the ATP-binding dwell. This is supported by a structure of mitochondrial F1 , thought to be poised at the ATP-binding dwell, that shows a rotary position nearly identical with that of ϵ-inhibited E. coli F1 [44,46]. Finally, recent biochemical studies of E. coli F1 confirmed that the ϵ-inhibited state is stabilized by Mg2+ADP and Pi but reversed by Mg2+AMP-PNP , consistent with the smFRET L*/M*/H* positions observed only with Mg2+ADP and Pi. Several bead-rotational studies with F1 from E. coli and other bacteria showed that ϵ inhibition pauses rotation for extended times, but concluded that ϵ pauses F1 at the catalytic dwell angle [47–49]. This contrast with the smFRET and structural results remains to be resolved.
Smaller step sizes of the rotary Fo motor
Driven by PMF during ATP synthesis, the step sizes of the c-ring with respect to the static a-subunit were smaller and revealed a one-proton-after-another mode of rotation in Fo according to smFRET . Using the geometric constraints of c-ring size and label positions, a 36° step size was most likely for about half of the assigned FRET level changes. Similarly, ten-stepped c-ring rotation was reported during ATP hydrolysis using immobilized FoF1 reconstituted in lipid nanodiscs with a gold nanorod as the marker of c-ring rotation .
Dwell times and rotational speed
The smaller step sizes in c-ring rotation during ATP synthesis were associated with shorter dwell times of the stopping positions . Measuring small dwell time differences with smFRET is possible, for example, the three slightly different catalytic dwell times for the ϵ-subunit indicated an asymmetry in rotation, eventually related to the asymmetric peripheral stalk affecting the conformational dynamics of the nearby nucleotide-binding site [33,35]. However, large changes in the dwell times were observed after addition of the non-competitive inhibitor aurovertin A, for the F1 as well as the Fo motor [34,51]. The inhibitor prolonged the dwell time during ATP hydrolysis and also resulted in a double-exponential decay with rise and decay components (Figure 2F). Dwell-time analysis has become an important control using inhibitors to discriminate conformational protein dynamics from single-molecule photophysical artefacts. However, time resolution limits for smFRET apply, by the binning of 1 ms for time trajectories and the difficulties to assign dwell times shorter than 5–10 ms from EFRET changes in noisy data.
Twisting and elastic energy storage with the rotor
SmFRET was also applied to detect a reversible elastic twisting mode within the rotor subunits ϵ and c of FoF1-ATP synthase during ATP hydrolysis and synthesis [52,53]. Transient elastic energy storage had been postulated to address the symmetry mismatch of the F1 and Fo motor step sizes and to ensure maximum efficiency of motor operation (experimental details are summarized in [2,54]). Using three different specifically attached fluorophores on a single FoF1-ATP synthase (EGFP–a fusion on the stator, Alexa Fluor® 532–ϵ and Cy5–c on the rotor), we could show that the distances between markers on residues ϵ56 and c2 fluctuated during rotor movement, indicating a twisting up to three single steps of c or 108° respectively .
smFRET of the CTD of ϵ
In this section, we present our preliminary development of smFRET to monitor conformational changes of the ϵ CTD in E. coli F1. Based on the E. coli F1 X-ray structure , we chose ϵ99 on the first C-terminal α-helix of ϵ, which does not insert into a β–γ cleft in the ‘up’-conformation (Figures 1B and 1C). The second marker position is γ108, yielding FRET distances of approximately 3 nm (Figure 1C) and 6 nm (Figure 1B) including 0.5 nm for linkers to the fluorophores to ϵ99 in the ‘up’ or ‘down’ conformations respectively. These labelling positions were also chosen to avoid perturbing any interactions of ϵ CTD (either conformation) with the ϵ NTD or with other subunits. This is in contrast with smFRET experiments of R. Iino and colleagues for the thermophilic enzyme TF1 from Bacillus PS3, in which both γ- and ϵ-labelling sites would be buried inside the F1 central cavity with ϵ in the ‘up’ state .
Our initial tests with smFRET probes were on freely diffusing F1 under different ligand conditions. Subunit ϵ was expressed separately with a His6 N-terminal affinity tag and was purified as before . A unique cysteine residue was included, and ϵ99C was labelled with Atto647N as FRET acceptor. Maleimide-labelling efficiency was 30%, and unbound dye was removed by dialysis. F1(γ108C), depleted of δ and ϵ , was labelled with Atto488 as FRET donor (maleimide-labelling efficiency 55%, unbound dye removed by centrifuge column). Mixing F1 (3 μM) with ϵ (4 μM) for 30 min yielded FRET-labelled F1, because of the high binding affinity of ϵ (Kd ~0.3 nM ). Dilution to less than 1 nM F1–ϵ immediately before starting smFRET measurements resulted in standard single-molecule detection conditions in solution for our confocal microscope, i.e. one F1–ϵ molecule at a time. Using alternating laser excitation with 488 nm for FRET between γ and ϵ, and 635 nm to probe the Atto647N-labelled ϵ bound to F1, allowed selection of the FRET-labelled enzymes, rejecting any protein aggregates or single-labelled proteins in subsequent analysis.
Diffusion of F1–ϵ (~10 nm diameter) was fast, i.e. approximately 3 ms on average through the confocal detection volume (against ~300 μs for a free fluorophore). These short observation times allowed us to determine only an average FRET distance for each enzyme, but not time-dependent distance changes or conformational changes between γ and the CTD of ϵ within a single photon burst. We obtained several hundred burst events with high photon count rates for each biochemical condition using the following thresholds to identify a single FRET-labelled F1–ϵ: a mean diffusion time longer than 10 ms, maximum peak intensity for the FRET donor fluorophore (to exclude aggregates with multiple dyes), fluorescence intensity thresholds for the FRET acceptor (at least a mean of four counts per ms for FRET excitation and eight counts per ms for direct excitation in the same photon burst) and limited FRET efficiency fluctuations of less than 0.18 (S.D. within a burst). Figures 3(A) and 3(B) show two photon bursts of FRET-labelled F1–ϵ in the presence of 1 mM Mg2+AMP-PNP. The FRET efficiencies (blue traces) show different average values, approximately 0.6 and 0.3, indicating different distances between the markers on γ108 and ϵ99, and corresponding to different conformations of the CTD of ϵ with respect to γ.
smFRET of ϵ conformational changes in F1
Addition of different ligand combinations in the presence of Mg2+ resulted in distinct EFRET distributions (Figures 3C–3E). The total number of FRET level in the three histograms depended on the photon burst criteria used to identify a single FRET-labelled F1–ϵ and therefore cannot be compared directly. Biochemical data showed that Mg2+ADP and Pi stabilize the ϵ-inhibited state . In Figure 3(C), addition of ADP/Pi resulted in a dominant population with EFRET of ~0.6, similar to the EFRET histogram obtained without adding nucleotides (results not shown). Given a Förster radius of 5.1 nm (Attotec) for EFRET=0.5, this corresponds to a 4.8 nm FRET distance and should represent the ϵ-inhibited ‘up’ state, as in the E. coli F1 structure  and in Figure 1(C). Adding AMP-PNP or ATP resulted in an additional population of EFRET about 0.25. This low EFRET value corresponded to a 6.1 nm distance between the FRET fluorophores and therefore should be the ‘down’ conformation of the CTD. However, the majority of F1–ϵ complexes were still found at EFRET of ~0.6. This probably correlates with the strong inhibition of isolated F1 by ϵ . The distance changes as calculated from the maxima of the two EFRET populations agreed with the changes seen in the structural models in Figures 1(B) and 1(C), but the absolute distances were larger than expected, which could be explained by possible photophysical effects of the local protein environment of the fluorophores, such as decreased quantum yields or spectral shifts. However, additional smFRET measurements are required to assign unequivocally the different FRET distances with the ϵ CTD conformations and its inhibitory role.
smFRET is a complementary approach to measure subunit rotation of the two motors in reconstituted single FoF1-ATP synthase. With a time resolution of 1 ms, dwell times of a few milliseconds for the stopping positions are accessible, and the angular resolution for the rotary movement can be inferred using known structural constraints of the enzyme. In addition, domain movements such as the conformational change of the regulatory CTD of ϵ can be monitored in real time.
In the present mini-review, we reported the nucleotide-dependent shifts in the population of the CTD between ‘up’ and ‘down’ states by smFRET of F1–ϵ in solution. Accordingly more than 50% of F1 on average remained in an inhibited ‘up’ conformation of ϵ in the presence of Mg2+ATP or Mg2+AMP-PNP, which is in agreement with video microscopy results of beads attached to immobilized F1 as a marker for rotation  and the role of PMF to activate the enzyme for ATP hydrolysis . We now need to reconstitute FRET-labelled F1 with Fo in liposomes to study the dynamics of the ϵ CTD conformations in the intact ATP synthase.
To improve smFRET-based analysis of the ϵ CTD, we have to increase the observation time for single enzymes in solution, using either a three-dimensional trap (e.g. the ‘anti-Brownian electrokinetic trap’, or ABELtrap, invented by Cohen and Moerner ) to hold the FoF1-liposome in place during smFRET recording, or integrating the FRET-labelled enzyme into a BLM (black lipid membrane) with access to single-molecule detection. The BLM approach allows one to control and change the PMF during the measurement . Furthermore, a three-fluorophore smFRET experiment will be important to correlate rotor movement and the conformation of the CTD of ϵ, and to minimize photophysical artefacts.
Interpretation of smFRET data requires structural information. More X-ray structures with atomic resolution will be important to advance our understanding of how the rotary motors and their controls operate in this enzyme. These data are also the basis for MD simulations of motors and controls that provide independent atomic views with high time resolution, but short ‘observation’ times in nanoseconds due to computational limitations. Structural information might elucidate the role of nucleotide (ATP) binding as possible part of the conformational dynamics of ϵ, and are essential to interpret the nucleotide-dependent binding constants of ϵ to F1. Our ongoing work on ϵ inhibition is now focusing on the complete enzyme reconstituted into liposomes, and will proceed to probe the regulatory conformational changes of ϵ and the rotary motors in the native environment of the E. coli enzyme, i.e. the plasma membrane of living cells.
Bioenergetics in Mitochondria, Bacteria and Chloroplasts: Third Joint German/UK Bioenergetics Conference, a Biochemical Society Focused Meeting held at Schloss Rauischholzhausen, Ebsdorfergrund, Germany, 10–13 April 2013. Organized and Edited by Fraser MacMillan (University of East Anglia, Norwich, U.K.) and Thomas Meier (Max Planck Institute of Biophysics, Frankfurt am Main, Germany).
We thank our co-workers M. Renz (Jena) and M. Hutcheon (Syracuse) for their excellent technical assistance for the smFRET measurements of the CTD of ϵ.
This work was funded by a collaborative grant from the National Institutes of Health [grant number R01GM083088-03S1 (to T.M.D.)] and in part by the Deutsche Forschungsgemeinschaft [grant number BO 1891/10-2 (to M.B.)].