Transport domain unlocking sets the uptake rate of an aspartate transporter.

Glutamate transporters terminate neurotransmission by clearing synaptically released glutamate from the extracellular space, allowing repeated rounds of signalling and preventing glutamate-mediated excitotoxicity. Crystallographic studies of a glutamate transporter homologue from the archaeon Pyrococcus horikoshii, GltPh, showed that distinct transport domains translocate substrates into the cytoplasm by moving across the membrane within a central trimerization scaffold. Here we report direct observations of these 'elevator-like' transport domain motions in the context of reconstituted proteoliposomes and physiological ion gradients using single-molecule fluorescence resonance energy transfer (smFRET) imaging. We show that GltPh bearing two mutations introduced to impart characteristics of the human transporter exhibits markedly increased transport domain dynamics, which parallels an increased rate of substrate transport, thereby establishing a direct temporal relationship between transport domain motion and substrate uptake. Crystallographic and computational investigations corroborated these findings by revealing that the 'humanizing' mutations favour structurally 'unlocked' intermediate states in the transport cycle exhibiting increased solvent occupancy at the interface between the transport domain and the trimeric scaffold.

Glutamate transporters, also termed excitatory amino acid transporters (EAATs), maintain glutamate concentration gradients across the cell membrane by coupling neurotransmitter uptake to symport of three sodium (Na+) ions and a proton and counter-transport of a potassium ion[1]. Structural information on the EAAT family principally stems from investigations of GltPh[2-6], an aspartate and Na+ symporter[7,8] from Pyrococcus horikoshii. Crystal structures of GltPh revealed that the homo-trimeric protein is composed of a rigid, central trimerization scaffold that houses three peripheral transport domains containing the substrate binding sites (). Comparison of GltPh structures captured in distinct conformations suggests that within the trimerization scaffold, individual transport domains undergo relocations approximately 15 Å normal to the membrane, providing substrate and ions alternating access to the extracellular (outward) and intracellular (inward) solutions ()[5].

Single-molecule imaging of GltPh provided direct evidence for ‘elevator-like’ transport domain motions[9,10]. Consonant with double electron-electron spin resonance (DEER) measurements[11,12], these measurements also showed that individual GltPh transport domains transition spontaneously between outward- and inward-facing conformations both when free of cargo (apo) and when bound to substrates. Notably, these transport domain motions exhibited heterogeneous dynamic behaviors, alternating between periods of rapid transitions and periods of quiescence[9]. In contrast to observations in structurally unrelated neurotransmitter sodium symporters[13], substrate binding decreased transport domain dynamics in GltPh by favoring the quiescent periods such that the frequency of domain motions converged to the substrate uptake rate[7,9].

These findings led to the hypothesis that GltPh configurations observed in crystal structures[2,4], showing tight lock-and-key interactions between transport and trimerization domains, represent quiescent “locked” states with high substrate affinity, whereas the short-lived states sampled during dynamic periods are structurally distinct and likely have intrinsically lower substrate affinity (). This model posits that transport domain motions require a rate-limiting, structural “unlocking” process that changes the interface between the transport and trimerization domains, likely enabling solvent penetration into that interface[5,9,14].

To assess the relationship between GltPh function, dynamics and structure, we employed smFRET imaging in the context of reconstituted proteoliposomes with physiological ion gradients. We compared wild-type (WT) GltPh to a gain-of-function, “humanized” (H) mutant R276S/M395R (H276,395-GltPh), which exhibits a faster rate of substrate uptake[15]. The smFRET experiments revealed that the mutations destabilized quiescent “locked” states. The resulting increase in dynamics paralleled a decreased affinity for substrate and an increased transport rate. Crystallographic analyses supported this observation, showing that the transport domains of H276,395-GltPh can adopt inward-facing conformations in which the transport domain-trimerization scaffold interface is dramatically more open than previously observed. Computational modeling further suggested that increased solvation by lipid or detergent hydrophobic tails in this interface likely facilitates the formation of such conformations. These observations provide a structural rationale for functional distinctions between GltPh and the human EAATs, and establish a kinetic framework for understanding how regulation can be achieved.

Experimental design

GltPh is a structural homologue of EAATs (~35% sequence identity)[2] that preferentially transports aspartate over glutamate, with higher substrate binding affinity and slower uptake rate[3,7]. Ryan and colleagues suggested[15] that these distinctions may stem, in part, from the differential location of a conserved arginine residue[16] that is proximal both to the substrate-binding site and the transport domain-trimerization scaffold interface. While the location of this arginine can differ in the primary sequences of glutamate transporter homologues, its position is conserved in most family members (). In human EAAT1, moving this arginine from trans-membrane segment (TM) 8 to helical hairpin (HP) 1 (where it is located in GltPh) markedly increases substrate affinity and decreases uptake rate[15]. Reciprocal mutagenesis of GltPh, whereby the arginine is moved from HP1 to TM8 (R276S/M395R), reduces aspartate affinity and increases the transport rate[15]. We took advantage of this gain-of-function mutant to probe correlations between uptake rate and transport domain dynamics.

We performed a comparative analysis of WT and H276,395-GltPh proteins using smFRET, where elevator-like transport domain motions in individual GltPh trimers, bearing one donor and one acceptor fluorophore, were revealed as time-dependent changes in FRET efficiency[9] (). To investigate such motions in the context of proteoliposomes, we labeled GltPh proteins with intra-molecularly stabilized Cy3 and Cy5 derivatives that exhibit intrinsically enhanced brightness and photostability[17,18] (), eliminating the need for fluorophore protective agents that disturb lipid bilayer properties[19]. The labeled proteins were reconstituted into liposomes in the absence of substrates for smFRET and bulk substrate uptake measurements. Bulk, radioactive aspartate uptake experiments confirmed that both labeled WT and H276,395-GltPh mutant transported substrate with rates similar to those of the unlabeled proteins, the mutant being about 4-fold faster than the WT ().

For smFRET measurements, reconstitution procedures were established to yield maximally 1 GltPh trimer per vesicle. Proteoliposomes were immobilized via biotinylated, fluorescently labeled GltPh within passivated quartz microfluidics chambers activated with a biotin-streptavidin bridge (). Using this strategy, only those proteoliposomes containing GltPh oriented with the extracellular side facing the vesicle exterior were immobilized and imaged[20]. Imaging experiments were initiated in the absence of substrates under isoelectric conditions and chemical gradients were established by rapidly exchanging the proteoliposomes into a buffer containing Na+ ions and aspartate. Additionally, we examined the behaviors of the labeled proteins in detergent micelles that afford increased statistics and higher signal-to-noise ratios.

Transport rate and dynamics are correlated

In both the absence and presence of gradients, WT GltPh in proteoliposomes showed spontaneous transitions between low-, intermediate- and high-FRET efficiency states centered at ~0.4, ~0.6 and ~0.9, respectively (). In detergent solutions, these FRET states were assigned to specific GltPh configurations: the low-FRET state reflects symmetric outward-facing and asymmetrically outward- and inward-facing configurations; intermediate- and high-FRET states reflect, respectively, asymmetrically inward- and outward-facing and simultaneously inward-facing protomers ()[9]. In line with previous investigations[9,11,12], population FRET data from hundreds of individual proteins in the absence of gradients show that the transporter occupies the outward-facing, low-FRET state about half of the time in both detergent (46%) and lipid vesicles (54%) (Fig. 2a, Extended Data Fig. 3b and Extended Data Table 1a-b). Transitions between low- and higher-FRET states reflect elevator-like movements of the individual transport domains between outward- and inward-facing configurations, respectively[9]. In proteoliposomes, such transitions occurred at a rate of ~0.2 s−1, roughly two-fold less frequently than in detergent (). Paralleling the effects of substrate binding to WT GltPh in detergent (), a modest population shift toward the outward-facing, low-FRET state occurred under active transport conditions achieved by addition of Na+ and aspartate (). Na+ and aspartate also reduced transport domain dynamics by 10-fold to ~0.02 s−1 (). Thus, in the presence of chemical gradients the frequency of transitions from outward- to inward-facing state (~0.01 s−1) mirrored the rate of radioactive substrate uptake (~0.007 s−1) ().

Figure 2

Ligand-dependent state distributions in detergent

In each panel, time-dependent FRET efficiency population contour plots (left) and cumulative population histograms (right) are shown for the WT (a) and mutant (c). Experimental conditions are indicated above the panels. Contour plots are color-coded from tan (lowest) to red (highest population);color scale from 0-12%. Histograms display the time-averaged state distributions. Solid black lines are fits to sums of individual Gaussian functions (red lines). N is the number of molecules analyzed. e, d, Corresponding transition density plots (as in Fig. 1).

Extended Data Figure 3

Conformational state distributions of WT and H276,395-GltPh in proteoliposomes

a, Examples of smFRET recordings. Top panels show raw fluorescent signals originating from donor (green) and acceptor (red) dyes. Bottom panels show changes of FRET efficiency calculated from raw data (blue). Red solid lines through the data are idealizations obtained using QuB software[35] b, Contour plots and one-dimensional population histograms in the absence and presence of Na+ and aspartate in the external liposome buffers. Buffer compositions inside and outside of the vesicles are shown above the panels. WT and H276,395-GltPh histograms are fitted to three and two Gaussian functions, respectively. c, Transitions density (TD) plots for the WT (left) and H276,395-GltPh (right) in proteoliposomes in the absence of Na+ and aspartate in the external buffer. d, Means and widths (in brackets) of FRET efficiency distributions derived from Gaussian fits to proteoliposome data in comparison to detergent data.

Extended Data Table 1

FRET state assignments and populations; time constants for the slow and fast components.

a, FRET State Population Distributions in proteoliposomes
WT Gltph
FRET	Subunit configuration	Apo, %		P(out)=0.55, %		Transport, %	P(out)=0.65, %
Low	OF/OF+OF/IF	54		55		63	65
Intermediate	IF/OF	27		25		22	22
High	IF/IF	19		20		15	13
R276S/M395R GItph
FRET	Subunit configuration		Apo, %		Transport, %
Lower	OF/OF+OF/IF		40		55
Higher	IF/OF+IF/IF		60		45
b, FRET State Population Distributions in detergent micelles
WT GItph
FRET	Subunit configuration	Apo, %		P(out)=0.45, %		Bound, %	P(out)=0.5, %
Low	OF/OF+OF/IF	46		45		49	50
Intermediate	IF/OF	24		25		25	25
High	IF/IF	29		30		24	25
R276S/M395R Gltph
FRET	Subunit configuration		Apo, %		Transport, %
Lower	OF/OF+OF/IF		62		30
Higher	IF/OF+IF/IF		38		70
c, Time constant for stable (slow) and transient (fast) FRET States in detergent micelles
WT	Low FRET					Intermediate / High FRET
	tFast, S		tSlow, S			tFast, S	tSlow, S
Apo	~ 0.6		~ 6			~ 0.6	~ 5
Na+, aspartate	~ 0.7		~ 12			~ 0.7	~ 15
R276S/M395R	Low FRET					Higher FRET
	t, s					t, s
Apo	~ 1.5					~ 1.1
Na+, aspartate	~ 1.7					~ 7.6

a-b, Shown are the assignments of FRET states to configurations of labeled subunit pairs and corresponding observed populations, rounded to integer numbers. Also shown are the calculated populations considering the probability of a protomer to be in the outward facing state P(out) and assuming independent protomers in the trimer. c, Time constants for the WT transporter, □□, of the slow and fast components were derived from fitting the survival data compiled from the measured dwell times to double exponential function. The time constants for the H276,395-GltPhmutant were obtained by fitting the survival data to a single exponential function. Shown are averages from three independent experiments. The standard errors are within 5%. Dwell times longer than 10 s are significantly underestimated because photobleaching, which occurs with time constant of ~40 s, is limiting the observation window.

Notably, the H276,395-GltPh mutant only exhibited transitions between low- (0.4) and a single, higher- (0.65) FRET state in both proteoliposomes and detergent (). Similarly to the WT protein in the absence of chemical gradients, the low-FRET state was occupied 60% of the time in detergent micelles and 40% in proteoliposomes (). The observed FRET transition frequency for H276,395-GltPh was also two times slower in proteoliposomes (~0.13 s−1) compared to detergent (~0.22 s−1) ().

In stark contrast to the WT protein, however, the transition frequency in H276,395-GltPh decreased by less than 2-fold to ~0.1 s−1 when transport-supporting chemical gradients were established (). Here again, the frequency of transitions from the outward- to inward-facing FRET state (~0.05 s−1) converged to the measured rate of substrate uptake (~0.03 s−1) (). The quantitative correspondence observed between the rates of smFRET transitions and uptake for the WT and H276,395-GltPh mutant proteins provides compelling evidence that elevator-like motions of transport domains mediate solute uptake and are critical steps of the transport cycle[9]. This finding was independent of the proteoliposome immobilization strategy used and valinomycin-mediated electrical potentials ()[10].

H276,395-GltPh samples a distinct inward-facing state

In contrast to WT GltPh, which samples intermediate- (0.6) and high- (0.9) FRET states, H276,395-GltPh samples only a single higher-FRET (0.65) configuration (). No excursions into the 0.9 FRET state were observed even when data were collected at 6-fold higher time resolution (15 ms) (). The absence of the 0.9 FRET state would be expected if only one protomer within the H276,395-GltPh trimer transitioned into inward-facing configuration at a time, while the formation of symmetric inward-facing states were disallowed. This model is, however, inconsistent with data showing that individual transport domains function independently[5,14,21,22]. An alternative hypothesis is that the inward/outward and inward/inward configurations in H276,395-GltPh exhibit altered, overlapping FRET values. If this model is correct, then the gain-of-function mutations in H276,395-GltPh have altered the nature of the elevator-like transport domain motions and the structure of the inward-facing state.

The energy landscape of H276,395-GltPh is altered

Na+ and aspartate significantly stabilized the higher-FRET state of H276,395-GltPh in detergent micelles (). In detergent, Na+ and aspartate have access to both the extracellular and cytoplasmic sides of the protein. Assuming that a binding equilibrium is established in each conformation, these observations suggest that substrates bind more tightly to the inward-facing H276,395-GltPh conformation. Such a response was not observed for the WT GltPh, where substrate affinities of the inward- and outward-facing conformations are nearly the same[23] and ligands stabilize the latter only slightly[9,11,12] (). Notably, the transporter blocker DL-threo-β-benzyloxyaspartate (TBOA)[24] stabilized the outward-facing low-FRET states of both WT and H276,395-GltPh (). As above, this suggests that TBOA preferentially binds to the outward-facing state of both isoforms. Results consistent with these findings were obtained from ensemble DEER measurements using the protein spin-labeled on the same residue ().

Interestingly, the addition of Na+ and aspartate to H276,395-GltPh proteoliposomes led to an increase in the outward-facing, low-FRET population (). This liposome-specific response to substrate addition provides supporting evidence for bilayer integrity. It also reveals that elevator-like transport domain movements –as opposed to substrate release –are rate-limiting in the H276,395-GltPh transport cycle. If substrate release were slow compared to the domain movements, the state distributions during uptake would mirror those observed in detergent, i.e. show preference for the inward-facing higher-FRET state.

The effect of substrate on the distribution of FRET states observed for both isoforms was concentration-dependent in detergent micelles. Notably, H276,395-GltPh exhibited an ~1000-fold increase in apparent substrate dissociation constant (K) compared to the WT protein (). This finding was corroborated by bulk measurements (). Hence, the H276,395-GltPh mutations affect both transport domain dynamics and substrate affinity even though neither of the mutated residues coordinates aspartate directly in the existing crystal structures. These observations support the hypothesis that substrate binding and transport domain dynamics are physically coupled.

Locked states are destabilized in H276,395-GltPh

The coexistence of quiescent and dynamic periods evidenced both in the absence and presence of ligands is a hallmark kinetic feature of WT GltPh[9]. Binding of Na+ and aspartate increases the prevalence of quiescent periods and thus the average FRET state lifetimes (). Strikingly, no evidence was found for quiescent periods in the H276,395-GltPh mutant () and rapid transport domain dynamics persisted even in the presence of saturating substrate concentrations (). These dynamic processes were efficiently blocked by TBOA () consistent with their putative role in substrate transport. In H276,395-GltPh, substrate binding increased the lifetime of the high-FRET state (~7-fold), with no detectable impact on the low-FRET state lifetime (). In both the absence and presence of ligands, the low- and higher-FRET state lifetimes were monodisperse (). These findings suggest that in H276,395-GltPh, the isomerization steps leading to locked configurations of the WT protein are dramatically altered or inaccessible under the conditions examined, although an allosteric coupling between substrate binding and stabilization of the domain interface still exists.

Structure of the inward-facing H276,395-GltPh

To probe the underpinnings of the altered properties of H276,395-GltPh, we determined a crystal structure of the protein bound to Na+ ions and aspartate at a moderate resolution of ~4.5 Å (). As expected from smFRET experiments (), the structural model clearly showed that all protomers in the trimer spontaneously adopted inward-facing configurations. The model also revealed that the transport domains orientations differed from those previously captured in GltPh structures, both with and without stabilizing cross-links[4,5]. Moreover, the trimer was asymmetric, with the transport domain of protomer A assuming a position distinct from the other two.

In protomer A, the transport domain shifted further inward by 2 Å and rotates by 7° around an axis roughly perpendicular to the membrane plane with respect to the WT (). This rearrangement is accommodated by a concerted movement of helices in the scaffold domain, comprising TM1 and peripheral portions of TM2 and TM5 (), whose flexible nature was already noted[5]. This conformation resembles the inward-facing, “locked” state of the WT[4,5] in the close packing observed between the transport domain and the trimerization scaffold (). Molecular dynamics (MD) simulations revealed that whereas Arg276 in the WT forms hydrogen bonds with Asp394 and bulk water molecules, the corresponding Arg395 in H276,395-GltPh faces the hydrophobic core of the bilayer. The resulting membrane remodeling is driven by the hydrophobic matching force[25,26], and is established by interactions of the Arg395 side chain with penetrating lipid phosphate groups and accompanying water molecules (). Consequently, the penetrating polar moieties are positioned in an otherwise hydrophobic region of H276,395-GltPh, which can destabilize the inward-facing, locked conformation and increase water accessibility to the substrate-binding site and to the domain interface ().

In protomers B and C the transport domains undergo identical and more dramatic changes (), each swinging away from the trimerization scaffold by about 12° compared to “locked” protomer A. Consequently, a large crevice opens between HP2 and the scaffold, reducing the interface between the transport and scaffold domains from ~1300 Å2 to ~900 Å2 and allowing access to water, detergent or lipid molecules (). This unusual, apparently “unlocked”, conformation was observed in two protomers occupying distinct crystal packing environments and therefore seems to be determined by the properties of the protein itself and not by crystal contacts. The crevice it generates is largely hydrophobic, and closes rapidly in MD simulations when solvated only by water (). In contrast, the open interface between transport and trimerization domains is stable with lipids positioned in this space (), suggesting that solvation by lipid or detergent molecules, is necessary. Notably, this crevice may allow HP2, whose gating role in the outward-facing state is well-established[3,6], to open when the transport domain is inward facing (). If so, the substrate release might be facilitated in the unlocked conformation, a notion compatible with the markedly reduced substrate affinity of this mutant.

Discussion

Conformational transitions between outward- and inward-facing states are key events in transport cycles of secondary active transporters[27,28]. In glutamate transporters and possibly other families[29-31], such transitions involve elevator-like movements of the substrate-binding domains supported by relatively rigid scaffold domains. The frequency of such transitions in GltPh in lipid bilayers and in the presence of physiological ionic gradients parallels the turnover rate of substrate uptake. This relationship also holds in a gain-of-function mutant H276,395-GltPh that exhibits a 1000-fold decreased substrate affinity and a 4-fold faster uptake rate. Collectively, our observations establish a direct correlation between the transport domain movements and substrate transport, and suggest an inverse relationship between substrate affinity and transport domain motions. The H276,395-GltPh mutant is special in this regard as other point mutations impact dynamics only and do not potentiate transport[9].

Importantly, the observed dynamic signatures strongly suggest that the rate-limiting step in this process is the unlocking of the transport domain from the trimerization scaffold (). While both the WT and the H276,395-GltPh proteins exhibit similar transport domain structures and translocate similarly positioned charged groups (including Arg276 in the WT and Arg395 in the mutant), locked states are relatively unstable in the H276,395-GltPh mutant, leading to overall faster dynamics and uptake.

The locked and unlocked configurations of WT GltPh, corresponding to quiescent and dynamic periods, respectively, coexist and interconvert spontaneously, which suggests that outward- and inward-facing states of GltPh –and by extension EAATs –should be viewed as structurally heterogeneous ensembles. Increased quiescent period durations in the presence of substrate further suggest that ligand binding is allosterically coupled to the formation of locked states[9]. Based on these insights, we propose a simplified kinetic framework for the transport cycle that recapitulates the most salient experimentally observed features (). The specific relationship of crystallographic snapshots of GltPh and related proteins to the topological features of this framework will need to be examined carefully.

The structure of H276,395-GltPh () captures an unlocked configuration that appears relevant to the proposed transport cycle and uniquely suitable for ligand binding and release. While the molecular basis of how the mutations in H276,395-GltPh impact the locked-unlocked isomerization requires further investigation, MD simulations suggest that protein-lipid interactions are pivotal (). The proposed role for the lipid hydrophobic tails in facilitating domain unlocking complements previous hypotheses that transient interface hydration facilitates transport domain translocation[5,14].

That two closely related GltPh isoforms exhibit distinct kinetic and structural signatures foreshadows the possibility that human EAATs differ substantially from GltPh, especially in their dynamic properties. Probing EAATs directly is therefore essential, particularly since the extent to which they might be diverted to kinetically stable, potentially off-pathway states may represent a regulatory modality.

METHODS

DNA manipulations, protein expression, purification and labeling

Single cysteine mutations were introduced by site-directed mutagenesis (Strategene) of a cysteine-less GltPh background, in which seven non-conserved residues had been replaced with histidines resulting in improved expression levels (termed GltPh from hereon for brevity)[3]. Constructs were verified by DNA sequencing and transformed into E. coli DH10-B cells (Invitrogen). Proteins were expressed as C-terminal (His)8 fusions as described previously[2]. Briefly, isolated cell membranes were re-suspended in Buffer A, containing 20 mM Hepes/NaOH, pH 7.4, 200 mM NaCl, 0.1 mM L-aspartate, 0.1 mM Tris(2-carboxyethyl)phosphine (TCEP). Membranes were solubilized in the presence of 40 mM n-dodecyl β-D-maltopyranoside (DDM) for 1 hour at 4°C. Solubilized transporters were purified by metal-affinity chromatography in Buffer A supplemented with 1 mM DDM and eluted in 250 mM imidazole. The (His)8-tag was cleaved by thrombin and proteins were further purified by size exclusion chromatography (SEC). For smFRET experiments, protein samples at 40 µM were labeled with a mixture of maleimide-activated Cy3 and Cy5 dyes that exhibit enhanced photostability[17,32] as well as biotin-PEG11, at concentrations of 50, 100 and 25 µM, respectively, for 30 minutes at room temperature. Labeled proteins were purified away from the excess reagents by size exclusion chromatography. Their purity and specificity of labeling were assessed by SDS PAGE, which was followed by fluorescence imaging and Coomassie staining.

Protein reconstitution into liposomes for smFRET analysis and transport assays

Labeled and unlabeled GltPh variants were reconstituted into liposomes as previouslydescribed[4,11]. Briefly, liposomes, prepared from 3:1 (w/w) mixture of E. coli total lipid extract and egg yolk phosphotidylcholine (Avanti Polar Lipids) in a buffer containing 20 mM Tris/HEPES, pH 7.4 and 100 mM KCl, were destabilized by addition of Triton X-100 at a detergent to lipid ratio of 0.5:1 (w/w). For reconstitution, proteins were added to lipids at final protein to lipid ratio of 1:1000 (w/w) and incubated for 30 minutes at room temperature. Detergents were removed by repeated incubations with Biobeads as described[11]. For smFRET and radioactive substrate uptake experiments, the same proteoliposomes were extruded through 100 nm and 400 nm filters, respectively. This reconstitution strategy yields at most 1 and 16 GltPh trimers per vesicle, respectively. Radioactive substrate uptake was measured as previously described[7]. Briefly, proteoliposomes were diluted into reaction buffer containing 20 mM Tris/HEPES, pH 7.4, 100 mM NaCl and 0.3 µM [3H] L-aspartate at room temperature. Aliquots were removed at appropriate times, diluted in ice-cold quenching buffer (20 mM Tris/HEPES, pH 7.4, 100 mM LiCl) and filtered through 0.22 μm filters (Millipore). Protein concentration was estimated by the absorbance at 280 nm after correcting for the fluorophore contributions to the value. The amount of substrate uptake was normalized per mole of GltPh monomers.

smFRET experiments

All experiments were performed using a home-built, prism-based total internal reflection fluorescence microscope constructed around a Nikon TE2000 Eclipse inverted microscope body using streptavidin-coated, passivated microfluidic imaging chambers[33]. Except when stated otherwise, labeled proteins (either detergent solubilized or liposome-reconstituted) were surface-immobilized via a biotin-streptavidin bridge. Except when stated otherwise, imaging experiments were performed in a buffer containing: 20 mM Hepes/Tris (pH 7.4), 5 mM BME, an enzymatic oxygen scavenger system comprising 1 unit/ml glucose oxidase (Sigma), 8 units/ml catalase (Sigma) and 0.1% glucose[34]. In addition, apo-GltPh experiments included 200 mM KCl, Na+/Asp-bound experiments included 200 mM NaCl and 0.1 mM aspartate and Na+/TBOA-bound experiments included 200 mM NaCl, 10 mM TBOA. For experiments in detergent micelles, the buffers were also supplemented with 1 mM DDM. For imaging under transport conditions, the experiments were initiated in the absence of substrates (apo condition) on both sides of the membrane and chemical gradients were established by rapidly exchanging the proteoliposomes into an uptake buffer containing 100 mM NaCl and 100 µM aspartate. All data were collected at an imaging rate of 10 s−1 (100 ms integration time), except when otherwise stated. Fluorescence trajectories were selected for analysis using custom-made software implemented in Matlab (Mathworks) according to the following criteria[30]: a single catastrophic photobleaching event; over 8:1 signal-to-background noise ratio; a FRET lifetime of at least 5 seconds. FRET trajectories were calculated from the acquired intensities, I and I, using the formula FRET = ICy5/(ICy3 + ICy5). Population contour plots were constructed by superimposing the FRET data from individual traces. Histograms of these population data were fit to Gaussian functions in Origin (OriginLab). The relative populations and dwell time distributions of each FRET state, as well as the transition frequencies between them, were obtained by idealizing the smFRET traces using QuB[35]. Transition density plots and the dwell time survival plots were plotted and fitted as described previously[13]. The logarithmic histograms of the dwell times were fitted to transformed probability density functions[9]. Over 300 molecules are included in each smFRET experiment to ensure that the experimental margin of error in the mean value of each distinct FRET state across the three experiments is less than 5%.

Crystallography

The R276S/M395R GltPh mutant was purified by SEC in buffer containing 10 mM Tris/HEPES, pH 7.4, 100 mM NaCl and 7 mM n-decyl-β-D-maltopyranoside (DM). Protein solution at 3.5 mg/ml was mixed at 1:1 (vol.:vol.) ratio with the reservoir solution, containing 50 mM sodium acetate, pH 5.6-6, 18–20 % PEG 400 and 100-150 mM magnesium acetate, and crystallized at 4° C by hanging-drop vapor diffusion. Crystals were cryoprotected in reservoir solution. Diffraction data were collected at National Synchrotron Light Source beamline X29. Diffraction data were indexed, integrated and scaled using the HKL2000 package[36]. Anisotropy correction was applied as described previously[5]. Further analyses were performed using CCP4 programs[37]. Initial phases were determined by molecular replacement in Phaser[38] using transport and trimeric scaffold domains as separate search models. The model was optimized by rounds of manual rebuilding in Coot[39] and refinement in Refmac5[40] with TLS[37]. During refinement, strict non-crystallographic three-fold symmetry constraints were applied to the three transport domains and to regions of the scaffold domain that are involved in trimerization interactions. In addition, strict two-fold symmetry constrains were applied to the entire B and C protomers, which exhibited identical positions of the transport domain. For the outward- and inward-facing states, published coordinates were used: accession number 2NWX[3] and 3KBC[4], respectively. For the open conformation of HP2 the accession number of the coordinates is 4OYF[6]. All structural renderings were generated using PyMol[41].

DEER measurements and data analysis

Measurements were performed at 60 K using a 17.3 GHz home-built Ku-band pulse spectrometer[42]. A standard four-pulse DEER sequence with π/2-π-π pulse widths of 16 ns, 32 ns and 32 ns, respectively, and a 32 ns π pump pulse was used routinely. The frequency separation between detection and pump pulses was 70 MHz. The detection pulses were positioned at the low-field edge of the nitroxide spectrum. The homogeneous background was removed from the raw time-domain signals and the distances were reconstructed from the baseline-corrected and normalized signals by using Tikhonov regularization method[43] and refined by maximum entropy method[44].

Molecular modeling

Molecular dynamics simulations using the Charmm27 force field (FF)[45] and updated lipid FF[46] were prepared as described previously[14] and run using the NAMD 2.9[47] software at 300K with PME electrostatics and standard parameters for the Charmm FF. Atomic coordinates for the inward-facing wild-type GltPh were taken from PDB entry 3KBC[4] Simulations with the Gromos 54a7 FF[48] were prepared using the LAMBADA / InflateGRO membrane-embedding protocol[49] and run with the Gromacs 4.6.1[50] simulation package with reaction-field electrostatics and standard cutoffs for the Gromos FF. All simulations included pure POPC membranes, except Charmm27 Trajectory 3 (), which contained a mixture of 18% POPC, 52% POPE, and 30% POPG (prepared with Charmm-GUI web-tool[51]), more similar to the composition of the liposomes used in experiments. In selected simulations (), backbone C-α atoms were subjected to harmonic restraint potentials centered on positions from the X-ray structure with a harmonic constant of 0.1 kcal/molÅ2 (NAMD) or 0.24 kcal/molÅ2 (Gromacs). Docking of detergent and POPC lipid molecules was performed with Autodock Vina[52] within the Chimera 1.8 visualization software[53]. Lipid insertion in Charmm27 Trajectory 3 was performed as follows: (i) a frame from the molecular dynamics trajectory after 48ns of simulation time was selected; (ii) several lipid molecules restricted to various regions of the interfaces in protomers A and C were docked, ignoring the water; (iii) docking poses among the highest ranked from all docking runs were combined, such that lipid molecules fill the available hydrophobic pockets without clashing with each other, and overlapping water molecules were discarded; (iv) locally minimization was performed with the Charmm27 force field, including solvent and side chains within 5 Å of inserted lipids; (v) the molecular dynamics simulation was restarted at 300K. Data processing and plots were performed in Matlab (Mathworks).

Kinetic simulations of smFRET data

For the simulations, we assumed that protomer motions are independent. The model presented in was employed to simulate the motions of individual protomers between outward- and inward-facing orientations in QuB[35]. The time-dependent configurations of two protomers were then assigned to FRET states as described (). FRET traces were generated at 100 ms time-resolution in Matlab (Mathworks) using a Gaussian distribution of FRET efficiency values and widths derived from our experimental data. Initial estimates of the kinetic parameters were based on exponential fits of the experimental dwell time distributions (). The parameters were then manually optimized to recapitulate the experimental observables[20]: population FRET histograms, TDPs and the dwell-time histograms ().

Assignment of FRET efficiency states

a, Shown are the crystal structures of GltPh trimers in symmetrical outward- (OF) and inward- (IF) facing states and a model of an asymmetric configuration with two outward- and one inward-facing protomers[2,4]. The structures are shown in surface representation and colored as in Extended Data Figure 1. Black lines connect Cα atoms of residue 378, and the corresponding distances are indicated above the structures. b, Expected FRET efficiency levels for these distances for all possible configurations of subunit pairs: outward/outward (OF/OF), outward/inward (OF/IF), inward/outward (IF/OF) and inward/inward (IF/IF)[4]. c, Intramolecularly stabilized 4S(COT)-maleimide Cy3 (n=1) and Cy5 (n=2) fluorophores used in this study synthesized as described previously[17,18] with the addition of two sulfonate groups for increased solubility.

Extended Data Figure 1

Elevator model of transport and spatial conservation of a positively charged residue in glutamate transporter family

a, GltPh protomers in the outward- (left) and inward-facing (right) conformation are shown in surface representation and viewed in membrane plane. Dashed lines represent an approximate position of the membrane hydrocarbon layer. In the inward-facing state, the transport domain (blue) is moved by ~15 Å across the bilayer relative to the trimerization domain (wheat). b, Schematic representation of dynamic mode-switching between stable and transient conformations. c, A single GltPh protomer is shown in cartoon representation. Cyan balls emphasize the amino acid positions at which potentially positively charged residues occur in glutamate transporter homologues. d, Occurrence frequencies of these residues at the marked positions (GltPh numbering). To obtain the frequencies, sequences were harvested from the PFAM database[54] (accession code PF00375). Sequences were parsed to exclude those with over 70 % identity and aligned using Clustal Omega[55].

Single-molecule dynamics using different liposome-attachment strategies and with higher time-resolution

a-d, Dynamic properties of H276,395-GltPh under transport conditions using a different surface-immobilization strategy and in the presence of electrical potential. a, Surface-immobilization strategy for proteoliposomes using His-tagged lipids. b, Transition frequencies for WT (top) and H276,395-GltPh (bottom) trimers reconstituted into his-tagged liposomes that were site-specifically labeled in just two protomers with intramolecularly photostabilized Cy3 and Cy5 fluorophores. c, A negative inside voltage potential was established in proteoliposomes by adding valinomycin to the uptake buffer. d, Transition frequencies for WT (top) and H276,395-GltPh (bottom) in the presence of valinomycin. Each experiment shown includes statistics based on > 250 individual molecules. The standard error in transition frequency measurements is approximately 0.015 s−1. e-f, Dynamic properties of H276,395-GltPh probed at 15 ms time resolution. Contour plots and one-dimensional population FRET efficiency histograms (e) observed for the humanized mutant in detergent solution in the absence (left) and presence (right) of 100 mM NaCl and 100 µM aspartate. Examples of single-molecule trajectories (f).

Population changes in response to ligand binding

a-b, TBOA binding to H276,395-GltPh measured in smFRET experiments. Contour plots and population FRET efficiency histograms in the presence of increasing concentrations of TBOA (a). Changes in low- (red) and high- (blue) FRET state populations as a function of TBOA concentration (b). Solid lines through the data correspond to the Hill equation y= ymin+ (ymax-ymin)(xn/(xn+Kn)) with K=2.4 mM and n=1. The data points shown are averages and standard errors from three independent biological replicates. c, Experimental time domain DEER data (left) and reconstructed distance distributions (right) for H276,395-GltPh (shown in colors) and WT transporter (black) spin-labeled on residue Cys378 in detergent solution. The data were collected in the absence of ligands (top), in the presence of 100 mM Na+ and 350 µM aspartate (middle) and in the presence of 100 mM Na+ and 480 µM TBOA (bottom). The red arrows above the distance distributions mark distances between residues 378 extracted from crystal structures of the symmetric outward- (OF/OF) and inward- (IF/IF) facing states. The data for the WT transporter were adapted from a published study[11]. Notably, the data show that in the apo transporter, outward- and inward-facing states are similarly populated. Binding of Na+ ions and aspartate favors the inward-facing state, while binding of TBOA favors the outward-facing state.

Aspartate binding experiments

a, FRET efficiency population contour plots determined for H276,395-GltPh in detergent micelles in the presence of 100 µM aspartate and increasing concentrations of Na+ ions (indicated above the panels). b-c, Representative aspartate binding isotherms derived from ITC experiments for the WT GltPh (b) and H276,395-GltPh (c) in the presence of 10 mM Na+ and 100 mM Na+, respectively. Notably, the binding of aspartate to H276,395-GltPh in the presence of 10 mM Na+ is too weak to measure (inset). Binding experiments were performed using small-volume Nano ITC (TA Instruments). Upper panels show raw data. The cell contained 30 μM (WT-GltPh) and 40 μM (H- GltPh) protein buffer containing 20 mM HEPES/Tris, pH 7.4 and 0.1 mM DDM and indicated concentrations of NaCl. The syringe contained Asp at 200 μM concentration in the same buffer; every injection contained 5 μl. Data were processed and analyzed using manufacturer’s software (lower panels). Solid lines through the data are fits to independent binding sites model with the following K, enthalpy (ΔH), and apparent number of binding sites (n): 380 nM, 15 kcal/mol and 0.65 for the WT transporter, and 285 nM, 16 kcal/mol and 0.68 for H276,395-GltPh.

Data collection and refinement for Na+ and aspartate bound R276S/M395R-GltPh

a, Table showing data collection and refinement statistics. Scaling and refinement statistics were obtained after anisotropy correction by ellipsoidal truncation using high-resolution cutoffs of 4.9 Å along the a and b axis, and of 4.2 Å along the c axis. b, Stereoview of the 2Fo-Fc electron density map for H276,395-GltPh contoured at 1.5 σ around residue Arg395 in unlocked protomer C. Protein backbone (maroon) is shown in cartoon representations and side chains are shown as lines and colored by atom type. c, Superimposed scaffold domains of the inward-facing WT and H276,395-GltPh are shown in cartoon representation. The labile portions are colored cyan (WT) and magenta (mutant). Helices bend at conserved Pro60 and Pro206 residues (spheres). d, Locked (left) and unlocked (right) mutant protomers viewed from the cytoplasm and shown in surface representation.

Arg395 adapts to its environment

a, The arginine side chain (Arg276 in the WT; Arg395 in H276,395-GltPh) is seen in MD simulations to engage in hydrogen bonding interactions. Shown is the extent of the hydrogen bonds formation as a function of simulation time in Charmm Trajectory 3 (see Extended Data Figure 10). The main interactions of the arginine in both mutant and WT are with water molecules, but the locations of the waters are very different. In H276,395-GltPh, Arg395 side chain is located 5 to 9 Å below the level of the membrane surface, so that the water molecules are those penetrating the membrane-protein interface due to remodeling the membrane. In the WT, the water molecules interacting with Arg276 are in the space created inside the protein. b, The minimum distance from WT Met395 (top) or mutant Arg395 (bottom) side chains to any lipid phosphate group (left) or any water molecule (right) in Charmm Trajectory 3. In H276,395-GltPh after the initial equilibration phase, lipid phosphate groups interact with Arg395 either directly (5 Å distance) or through water (7.5 Å distance). In the WT, lipid head groups remain far from the hydrophobic Met395 side chain. Water interacts constantly with Arg395, but only occasionally with Met395 (in protomer B, a water molecule approaches M395 from the inside of the protein, at the interface between transport and trimerization domains). c, The same set of distances as in (b) for the mutant, from a different trajectory (G54a7 Trajectory 2) obtained independently, utilizing a different force field. The same trends are observed as in (b), showing proximity to the polar environment. d, Membrane bending (blue indicates thinning, red indicates thickening) close to Arg395 (green) which exposes its side chain to a polar environment comprised of water molecules and lipid head groups. e, RMSD of the Arg395 side chain with respect to the crystal structure after alignment on the trimerization domain, calculated from Charmm Trajectory 3 and G54a7 Trajectory 2. The side chain initially samples different conformations before settling into the membrane-exposed position shown in panel (d).

Extended Data Figure 10

Simulated smFRET data recapitulate experimental observations

a-d, Simulated FRET efficiency population contour plots (left side of each panel) and cumulative population histograms (right side) for WT GltPh (a) and H276,395-GltPh (b), and the corresponding transition density plots (c and d), (see for corresponding experimental data). As noted before[9], there are fewer transitions observed between the low- and high-FRET states in the WT transporter than would be expected from the model. This may either be because the model does not recapitulate the noise correctly or it may reflect previously uncharacterized communication between the protomers that warrants further investigation. e-f, Dwell time distributions for the low- (left panel) and intermediate- and high-FRET states (right panels) obtained for WT GltPh (e) and H276,395- GltPh (f) (see for corresponding experimental data).

Lipids or detergent molecules stabilize the unlocked conformation of H276,395-GltPh

a-e, Center-of-mass distance between the transport and scaffold domains of protomers A, B, and C of H276,395-GltPh as a function of MD simulation time.. The data are from five independent simulations initiated with position restraints on the Cα atoms (later released at different time points) and with the domain interface solvated with water. The vertical green lines indicate the moment in the corresponding trajectory when position restraints were turned off. Panels a and b show two repeats of the same starting structure simulated with the Charmm force field[45] and panel c with Gromos force field[48]. The transport domains in protomers B and C collapse onto the trimerization domain rapidly and loose their ligands in some cases (red arrows). d, A simulation, in which lipid tails partially insert into the interface spontaneously; the unlocked structure is stable much longer (note the different time scales on the time axis), and the collapse is only partial. e, The trajectory of a NAMD simulation (Charmm force field) in which lipid molecules were docked into the interface of protomers B and C at the time marked by the red arrow (3 lipids per protomer). The lipids remained in the docked region for the entire duration of the simulation and stabilized the position of the transport domain. f and g, The best scored docking poses for a detergent molecule and a POPC lipid, respectively, docked at the interface of protomer C.