For our study, we produced wild-type protein GlnBP (GlnBP WT) and two double-cysteine variants for analysis of conformational states via smFRET: GlnBP(111 C-192C) with point mutations at V111C and G192C (Figure 2—figure supplement 1) and GlnBP(59 C-130C) with point mutations at T59C and T130C (the latter was adapted from Feng et al., 2016; Zhang et al., 2020; Figure 2—figure supplement 1C). All protein variants were expressed in E. coli and purified using affinity chromatography (see Materials and methods for details). Protein purity was assessed by Coomassie-stained SDS-PAGE analysis (Figure 2A). As reported previously, GlnBP co-purifies with bound glutamine (Tjandra et al., 1992), which was removed by unfolding and refolding of the purified protein. We verified the monomeric state and proper folding of the resulting protein using size-exclusion chromatography (SEC, Figure 2B) by comparing the elution volume and shape of the monodisperse peak of GlnBP before and after the procedure (Figure 2—figure supplement 2).

To assess the binding affinity of GlnBP WT and the two GlnBP cysteine variants for L-glutamine, we performed ITC (Velázquez-Campoy et al., 2004). Refolded GlnBP WT showed a K_d for L-glutamine of 22±7 nM (Figure 2—figure supplement 3A) and K_d values of 31±3 nM and 35±5 nM for the two cysteine variants (Figure 2C, Figure 2—figure supplement 3B). These values are in overall agreement with previously published data (Chen et al., 2020). This verifies that the unfolding and refolding process as well as cysteine substitutions did not impact the biochemical properties of unlabeled GlnBP.

After assessing the thermodynamic properties of GlnBP, we characterized the conformational states and changes associated with ligand binding via smFRET. With smFRET, it is possible to study biomacromolecules in aqueous solution at ambient temperature, and identify conformational changes, heterogeneity, small sub-populations, and determine microscopic rates of conformational changes (Lerner et al., 2021; Lerner et al., 2018; Hellenkamp et al., 2018). We performed smFRET experiments on freely diffusing (Figure 3) and surface-immobilized GlnBP using the refolded variants GlnBP(111C-192C) and GlnBP(59C-130C) labeled with two different dye pair combinations, AF555/AF647 and ATTO 532/ATTO 643, to assess any position- and fluorophore-dependent effects. The smFRET assays were designed such that the inter-dye-distance of the apo state results in a lower FRET efficiency as compared to the holo state of the protein (Figure 2—figure supplement 1A/C).

Solution-based µsALEX (Lerner et al., 2018) data of GlnBP(111 C-192C) labeled with AF555/AF647 are shown in Figure 3B after an all-photon burst search (Nir et al., 2006). Both apo and holo states, in the absence and presence of saturation levels of glutamine, respectively, show a clear predominant population of donor-acceptor-labeled protein at S*-values of ~0.5, with two distinct mean apparent E* values for the apo (mid FRET, 0.51) and holo (high FRET, 0.68) states (Figure 3B/C, Figure 3—figure supplement 1A). This can be interpreted as a transition from open (apo) to closed (holo) GlnBP conformations upon the addition of the ligand. Similar results were obtained for the second double-cysteine variant (GlnBP(59 C-130C), Figure 3D, Figure 3—figure supplement 2) and from measurements with a different pair of fluorescent dyes for GlnBP(111 C-192C) (ATTO 532/ATTO 643; Figure 3—figure supplement 1B). Notably, further analysis and a comparison of mean accurate FRET efficiencies and the inter-dye distances show good agreement with simulated inter-dye distances of ±0.2 nm using the open- and closed-GlnBP crystal structures except for the holo-state of GlnBP(59 C-130C), which deviated by ~0.8 nm (Figure 2—figure supplement 1B/D).

Importantly, a quantitative analysis of the fraction of the closed state (high-FRET) subpopulation as a function of ligand concentration (Figure 3C and D) with a simple binding isotherm and no cooperative effects (n=1) provides K_d values in the 20–50 nM range for all labeled GlnBP variants. These results are fully consistent with ITC (Figure 2C, Figure 3—figure supplements 1/3). Interestingly, we found that arginine, the non-cognate ligand of GlnBP, induces hardly any FRET shifts even at millimolar concentrations of ligand (Figure 3—figure supplement 4) despite its binding to GlnBP at these concentrations (Figure 3—figure supplement 5).

We were also unable to identify a clear high-FRET subpopulation in the absence of a ligand, ruling out intrinsic conformational dynamics on timescales slower than the burst duration, >10 ms (Figure 3B, apo). To estimate an upper bound of the fraction of poorly sampled low abundance states, we determined the percentage of bursts outside of the main FRET population in the range of <E*>±σ of the characteristic FRET population of apo GlnBP(111 C-192C). For this, the FRET populations were fitted with Gaussian functions (with mean values and σ), which serves as a good approximation for mean E* values. For representative data sets of AF dyes, we found ~12% bursts outside of the main peak range (E*_holo = 0.64, σ=0.061, N=626; E*_apo = 0.47, σ=0.070, N=5,013) and for ATTO dyes, ~4% bursts outside the peak region (E*_holo = 0.56, σ=0.056, N=124; E*_apo = 0.37, σ=0.047, N=2,908). This suggests an upper bound of 5–10% for a subpopulation of other FRET subpopulations, for example partially closed conformations of GlnBP. Thus, our results agree with the idea that GlnBP mainly exists in one state – the open conformation – in the absence of its ligands.

Next, we analyzed our smFRET data for ‘within-burst’ dynamics using burst-variance analysis (BVA; Torella et al., 2011), multi-parameter photon-by-photon hidden Markov modeling (mpH²MM; Harris et al., 2022), intensity-based FRET efficiency versus donor lifetime (E−τ\begin{document}$E-\tau$\end{document}; E stands for FRET efficiency, τ\begin{document}$\tau$\end{document} is lifetime) plots (Kalinin et al., 2010) and burst-wise fluorescence correlation spectroscopy (FCS). These analyses provide access to FRET-dynamics that occur on timescales from a few milliseconds down to the sub-µs regime. This allows us to assess whether the observed FRET populations represent stable conformational states or time averages of (rapidly) interconverting states.

We first performed BVA of GlnBP(119-192) data with ATTO 532/ATTO 643 as a dye pair using a dual-channel burst search (DCBS; Nir et al., 2006). In BVA, within-burst E*-dynamics are identified as an elevated SD of the apparent FRET efficiencies, σ(E*), beyond what is expected from photon statistics, that is σ(E*) values larger than the theoretical semicircle (Figure 4—figure supplements 1–3, panels A). Our analysis indicates that, for each of the different ligand concentrations, at least some of the recorded single molecules undergo dynamic changes in E* while diffusing through the confocal spot (Figure 4—figure supplement 1). The within-burst dynamics are more prominent for dyes AF555 and AF647 (Figure 4—figure supplement 2) and become most abundant in the variant GlnBP(59C-130C), which was used in previous studies (Feng et al., 2016; Zhang et al., 2020; Wu et al., 2022; Figure 4—figure supplement 3). It is important to note that dynamic changes in apparent FRET efficiency can have photophysical origins and do not necessarily confirm the presence of conformational dynamics. For example, the apparent dynamic changes in E* might represent within-burst dynamics between FRET-active subpopulations (i.e. S*~0.5) and FRET-inactive subpopulations (e.g. donor-only, acceptor-only). Therefore, it is essential to quantify the BVA observed dynamics and identify the corresponding E*-S* subpopulations between which the dynamic transitions occur.

For this purpose, we used multi-parameter photon-by-photon hidden Markov modeling (mpH²MM; Harris et al., 2022; Pirchi et al., 2016) to identify the most-likely state model that describes the experimental results based on how E* and S* values change within single-molecule bursts. Such analysis can provide rates of exchange between distinct states of E*/S* and its interpretation is described in detail in Appendix 1. The mpH²MM analyses can differentiate whether apparent dynamic changes in E* arise from two conformational subpopulations or from photophysical transitions that do not represent conformational dynamics of GlnBP. Our analysis in Figure 4B shows clear signatures for donor- and acceptor-blinking between bright and dark states of the fluorophores (Figure 4, Figure 4—figure supplements 1–3), that is the FRET species with intermediate S* exchange with species of very high and low S* values, respectively. mpH²mm identifies a single and static apo FRET-active mid-E* state in the absence of a ligand and a single and static FRET-active high-E* state in the presence of saturating levels of ligand, which we ascribe to the open (mid-E*) and closed (high-E*) conformations of GlnBP. It is only in the presence of low concentrations of glutamine (around its K_d) where two FRET-active subpopulations, representing two distinct conformational states, are identified that might interconvert on timescales slower than 10ms (i.e. slower than typical burst durations). In conclusion, if intrinsic conformational dynamics existed in apo or holo GlnBP, it could only be between the highly populated FRET conformation we identify and another conformation that is populated significantly below the sensitivity of our measurement and analysis (i.e. a minor population with a fraction <5–10%) or these transitions would have to occur much faster than the time resolution of our experiments (<100 µs), which is dictated by the alternation periods in the µsALEX experiment.

To check for the presence of faster dynamics, we used multiparameter fluorescence detection with pulsed interleaved excitation (MFD-PIE) (Kudryavtsev et al., 2012). GlnBP(119-192) labeled with ATTO 532/ATTO 643 were used since this combination showed the least photophysical artifacts. In Figure 5, we first show two-dimensional plots of FRET efficiency (E) versus donor fluorescence lifetime values in the presence of acceptor (τD(A)\begin{document}$\tau_{D(A)}$\end{document}) for apo and holo GlnBP. The theoretical linear relationship between E and τD(A)\begin{document}$\tau_{D(A)}$\end{document} defines the static FRET line (Figure 5A, black lines). When the labeled molecules exhibit dynamics faster than the diffusion time, the fluorescence-weighted-average of the donor lifetime becomes biased towards longer donor lifetimes due to the higher brightness values of low-FRET species (Kalinin et al., 2010). Therefore, fast conformational switching is seen as bursts with distinct FRET efficiency values exhibiting a population shift towards the right of the static FRET line. As can be observed from the E−τ\begin{document}$E-\tau$\end{document} plots (Figure 5A), the center of mass of the FRET populations for both apo and holo GlnBP are coinciding with the static FRET line, suggesting the absence of conformational changes on timescales faster than milliseconds in line with data in Figure 4.

We also looked for dynamics using burst-wise FCS analysis (Figure 5B). For this, bursts containing signal from both fluorophores were selected, padded with 50 ms before and after burst identification, and the fluorescence autocorrelation functions of donor (Figure 5B, green curves) and acceptor signals (Figure 5B, red curves), as well as for the fluorescence cross-correlation functions between donor and acceptor signals (Figure 5B, black curves), were calculated. Conformational dynamics are expected to manifest themselves as an anticorrelation contribution in the cross-correlation function between donor and acceptor channels due to fluctuations in FRET efficiencies that occur faster than the translational diffusion component of the correlation functions (~1 ms on our setup; Torres and Levitus, 2007). The burst-wise FCS analysis at times <100 µs resulted in plateaued cross-correlation functions (Figure 5B, black lines) for apo and holo states, indicating the lack of dynamics down to the time-resolution of the experiments, that is the typical clock time of the photon time tagging on the order of 100 ns. It has to be mentioned that minor population exchange concerning <10% of molecules cannot be excluded with absolute certainty, particularly for the time regime <10 µs. Here, the noise increases due to limited photon budget, yet no clear indication for a cross-correlation related to conformational changes is seen, also supported by non-systematic fluctuations in the residuals (Figure 5).

Next, we characterize GlnBP and its conformational dynamics on timescales beyond the residence time of molecules in the confocal excitation volume (i.e. >1–10 ms) with the hope to obtain information on rare conformational events. We consequently conducted smFRET with NTA-based surface-immobilization of the GlnBP His-tag using TIRF microscopy (see Appendix 2 and accompanying Appendix 2—figures 1–5 for details). We reasoned that this would also allow the direct comparison of our results to those of Wang, Yan, and co-workers (Feng et al., 2016; Zhang et al., 2020; Wu et al., 2022). Importantly, in our analysis, we found that various buffer additives used for oxygen depletion have the same effect on GlnBP as the addition of glutamine (i.e. apo-GlnBP becomes artificially ‘closed’ in the presence of the additives) as we demonstrated in solution-based µsALEX experiments (Appendix 2—figure 1). Consequently, these additives were omitted since their effects mimic that of substrate binding. Strikingly, the conformational states of GlnBP were also partially altered upon surface immobilization (Appendix 2—figure 2), that is the E* values of GlnBP in apo/holo-state were significantly higher than in solution (Appendix 2—figures 2 and 4). Furthermore, GlnBP did not retain its full biochemical activity on the glass coverslips (Appendix 2—figure 2), that is only ~50% of all GlnBP molecules showed the expected shift towards higher FRET values upon addition of the ligand (Appendix 2—figure 2F). To validate our setup and immobilization approach, we additionally tested dsDNA (Appendix 2—figure 2A,C) and the two previously studied proteins SBD1 and SBD2 (Appendix 2—figure 5). Here, we did not observe discrepancies in FRET efficiency or biochemical activity, and the data of freely diffusing and surface-immobilized species were consistent.

Our combined smFRET analysis of GlnBP under different biochemical conditions suggests that conformational changes are tightly coupled to the ligand glutamine (Figure 3). We can also rule out prominent conformational dynamics on timescales between 100 ns and 10ms of apo and holo GlnBP via mpH²MM, MFD-PIE, and burst-wise FCS (Figures 4 and 5). Furthermore, our analysis suggests that apo GlnBP does not adopt (partially) closed conformations on the timescale >10 ms with an abundance >5–10% (Figures 3—5). While these results provide valuable information on ligand-binding affinity, conformational heterogeneity, and timescales of conformational dynamics in GlnBP, they are insufficient to exclude one or the other kinetic ligand-binding mechanism (IF vs. CS). We thus decided to integrate the obtained information into a general theoretical framework for analysis of ligand-binding mechanisms (Paul and Weikl, 2016; Weikl and Paul, 2014; Vogt and Di Cera, 2012), for which additional knowledge of the association and dissociation rates of ligand binding are required.

Such kinetic information is available from SPR spectroscopy or stopped-flow experiments. Since SPR was available to us, we immobilized GlnBP via its His-tag on a sensor chip and monitored its interaction with glutamine as a function of time. Even though GlnBP became partially inactive during immobilization for smFRET in TIRF microscopy (Appendix 2—figures 1–5), we reasoned that non-functional GlnBP will not be observed in SPR since only functional protein can contribute to the signal changes. The assumption that GlnBP remains functional on SPR-chips was validated by the match of ligand-binding characteristics obtained from ITC (Figure 2C, Figure 2—figure supplement 3), smFRET (Figure 3C and D), and SPR (Figure 6A).

In SPR, GlnBP showed specific and stable interaction with glutamine based on the magnitude of the equilibrium RU response as a function of glutamine concentration (Figure 6A). Analysis of the concentration-dependent maximal RU units yields a K_d of 10 nM (Figure 6A). The overall maximal response of around 3–4 RU indicates a 1:1 stoichiometry of glutamine assuming a monomeric state of GlnBP (Figures 2C and 6). Kinetic association and dissociation experiments were conducted under pseudo-first order conditions, that is the assumption of constant glutamine concentrations during an SPR run, due to the applied flow of buffer. The data were analyzed with the standard two-step reaction scheme (Schuck and Zhao, 2010; Schasfoort, 2017):(1)Glnbulk⇌ktktGlnsurface +GlnBP⇌koffkonGlnBP-Gln\begin{document}$$\displaystyle \text{Gln}_{\text{bulk}} \xrightleftharpoons[k_{t}]{k_{t}} \text{Gln}_{\text{surface}}\ +\text{GlnBP} \xrightleftharpoons[k_{\rm off}]{k_{\rm on}} \text{GlnBP-Gln}$$\end{document}

This includes a mass-transport step between the bulk solution of the applied flow and the sensor surface with transport rate k_t in both directions, and a binding step with effective on- and off-rate constants, k_on and k_off. Because of the dominance of mass transport, fits of this reaction scheme to the SPR sensorgrams using fit parameters k_t and k_on (after substituting k_off with K_d k_on in the scheme) do not allow determination of k_on within reasonable error bounds. However, fits with fixed values of k_on indicate that effective on-rate constants smaller than 10⁷ M^–1s^–1 are incompatible with the sensorgrams (Figure 6B,C, Figure 6—figure supplement 1). More precisely, plots of the rescaled sum of squared residuals for these fits versus k_on (Figure 6D) indicate a lower bound of at least 3⋅10⁷ M^–1s^–1 for k_on; this implies k_off = K_d k_on>0.3 s^–1 (with K_d = 10 nM). Among the 13 plots in Figure 6D, and among the 4 plots for [Gln]=125 nM, only 1 plot exhibits a minimum of the sum of squared residuals below this bound and is therefore likely an outlier.

To further investigate the coupling between conformational changes of GlnBP and ligand binding/unbinding, we performed atomistic simulations starting from the ligand-bound GlnBP structure with the AMBER20 software implementation for graphics processing units (GPUs; Case et al., 2020; Salomon-Ferrer et al., 2013) and the ff14SB force field parameters (Schuck and Zhao, 2010; Maier et al., 2015; see Materials and methods for details). To observe unbinding events on the microsecond timescale accessible in our simulations, we reduced all interactions between the protein and the ligand by 16%. With these reduced interactions, we observed ligand unbinding and a conformational change from the closed to the open GlnBP conformation in 5 out of 20 simulation trajectories with a length of 2 µs. Figure 7 illustrates characteristic distances for GlnBP opening and ligand unbinding on these 5 trajectories for time windows of 500 ns around the unbinding point. GlnBP opening is monitored by the distances between the C-α atoms of the residues 117 and 137 in domain 2 and the residue 51 in domain 1. We chose these residue pairs because they exhibit large relative changes in distance, with distances of 4.5 and 7.5 Å between residues 51 and 117 and residue 51 and 137 in the closed GlnBP conformation, respectively, and distances between about 15 and 30 Å in the open conformation for both pairs. Ligand unbinding is monitored by the root mean square deviation (RMSD) between the non-hydrogen atoms of the ligand in the simulation structures and the ligand in the bound crystal structure, after alignment of either the D1 or the D2 protein domain of these structures. These two RMSDs quantify the distance of the ligand to its native binding position on D1 and D2, respectively.

In trajectories 1, 2, and 4 in Figure 7, ligand unbinding occurs clearly after the opening transition of the protein, in agreement with the IF pathway of Figure 1C. During the opening transition of these trajectories, the ligand RMSD to the native binding position on D2 increases, which reflects the breaking of the ligand contacts to D2 during opening. The ligand RMSD to the native binding position on D1 remains low until the unbinding point, at which also the ligand contacts to D1 break. On the trajectories 3 and 5, in contrast, the ligand already unbinds during the opening transition of the protein, but also only after substantial opening at distances d_ub of the residues 117 and 137 to residue 51 at the unbinding point that are much larger than the corresponding distances in the bound conformation. It is important to note that the reduction of the protein-ligand interactions in our simulations lowers the binding free energies of the two protein conformations rather homogeneously, akin to a reduction of the ligand concentration, and reducing the ligand concentration is known to shift the flux towards the conformational-selection route of Figure 1C (Weikl and von Deuster, 2009; Hammes et al., 2009; Michel, 2016), if parallel pathways are possible.

Based on the simulations, IF seems the dominant binding mechanism also for the original, un-rescaled protein-ligand interactions of our atomistic model, because the sequences of opening and unbinding events observed on our simulations at weakened interactions clearly point towards IF, and because the weakening of the interactions rather decreases than increases the tendency for IF in Figure 1C. For the original, un-rescaled protein interactions, we expect significantly longer dwell-times in the closed state, significantly longer times for ligand unbinding from D1 after domain unbinding compared to the trajectories 1, 2, and 4 of Figure 7, and a significantly lower probability for ligand unbinding already at protein opening as on trajectories 3 and 5.

In addition, we performed simulations starting from the closed GlnBP structure with removed ligand to explore the conformational dwell times in the exchange between the closed and open conformation in the ligand-free state. We observed transitions from the closed to the open GlnBP conformation on 11 out of 20 MD simulation trajectories with a length up to 3 µs (see Figure 7—figure supplement 1). On the remaining nine trajectories, the closed conformation persisted for the simulation length of 3 µs. The fraction P(t) of trajectories that exhibit an opening transition up to timepoint t points towards a mean dwell time of several hundred nanoseconds for the closed conformation in the ligand-free state (Rhee et al., 2004). On the 11 trajectories that exhibited opening transitions, no subsequent transitions back to closed conformation were observed, which indicates clearly longer dwell-times in the open ligand-free GlnBP conformation of the simulations.

Two GlnBP double cysteine variants were generated by site-directed mutagenesis, allowing the insertion of two cysteine residues into GlnBP at positions (V111C – G192C) and (T59C – T130C), separately. E. coli BL21-pLysS cells were freshly transformed with the plasmid carrying the coding sequence for GlnBP WT or a GlnBP variant, and grown in 2 L LB medium (100 mg/mL Kanamycin and 50 mg/mL chloramphenicol) at 37 °C under aerobic conditions. At an OD_600nm of 0.6–0.8, overexpression of the proteins of interest was induced upon addition of 1 mM IPTG to the culture media. The cells were further grown for 1.5–2.0 hr after induction and then harvested by centrifugation for 20 min at 1529 × g (Beckman, JA10) at 4 °C. All subsequent operations were carried out at 4 °C, and all solutions were stored at 4 °C. Cell pellets from 2 L culture were collected in a 50 mL falcon and resuspended in buffer A (50 mM Tris-HCl, pH 8.0, 1 M KCl, 10 mM imidazole, 10% glycerol) with 1 mM dithiothreitol (DTT) and gently shaken overnight at 4 °C.

Cells were disrupted by sonication (Branson tip sonication; amplitude: 25%; 10 min; 0.5 s on-off pulses; temperature was kept low by the use of an ice-water bath). Centrifugation was used to fractionate the cell lysate (at 4 °C for 30 min at 4416 × g, Eppendorf, Centrifuge 5804 R) and at 4 °C for 1 hr for ultracentrifugation (70,658 × g, Beckman, Type 70Ti) in vacuum, and the pellet was discarded. The protein was purified by affinity chromatography using the Ni²⁺-Sepharose fast flow resin (GE Healthcare), pre-equilibrated with 10 column volumes of buffer A containing 1 mM DTT and gravity loaded with the supernatant from the preceding ultra-centrifugation step. The resin-bound protein was washed with 10 column volumes of buffer A containing 1 mM DTT, followed by buffer B containing 1 mM DTT (50 mM Tris-HCl, pH 8.0, KCl 50 mM, imidazole 20 mM, glycerol 10%), and finally eluted in buffer C (50 mM Tris-HCl, pH 8.0, KCl 50 mM, imidazole 250 mM, glycerol 10%) with 1 mM DTT. The eluted sample was concentrated (Vivaspin6 columns, 10 kDa MWCO, 6 mg/mL), dialyzed against PBS buffer supplemented with 1 mM DTT, and stirred gently at 4 °C overnight. SDS-PAGE was used to quantify the yield of protein overexpression and purification (Coomassie staining). The absorbance at 280 nm was used to estimate the protein concentration (knowing the molar extinction coefficient of GlnBP ~25,900 M^–1 cm^–1). The protein was then split into aliquots and kept at –20 °C. All proteins were further purified using SEC (ÄKTA pure system, Superdex 75 Increase 10/300 GL, GE Healthcare). The purified protein was split into aliquots and stored at –80 °C prior to the measurements.

The stock concentrations of GlnBP variants were estimated at about 6 mg/mL. Each GlnBP variant was thawed from –80 °C, then the protein was diluted to a final concentration of 3–4 µM (final volume of ̴20 mL) in the unfolding buffer (PBS buffer) containing 6 M guanidine hydrochloride (GndHCl). Subsequently, the solution was incubated for 3 hr under gentle stirring at ambient temperature. Next, the unfolded GlnBP variants were centrifuged (3046 × g, 30 min at 4 °C) to remove insoluble aggregates which could act as nuclei to trigger aggregation during the refolding process. A Snakeskin TM dialysis membrane was prepared (pre-cooled at 4 °C and soaked in refolding buffer - PBS buffer with 1 mM DTT, pH 7.4 - for 2 min). The GlnBP variants were transferred into the dialysis tubing, which were sealed tightly afterwards by double-knots and clips at each end. The unfolded GlnBP variant was refolded by a two-step dialysis, in the presence of a total 200-fold excess of refolding buffer. First, each protein was dialyzed against 2 L refolding buffer overnight under gentle stirring at 4 °C. Then, buffer was exchanged with an additional 2 L refolding buffer for another day at 4 °C. The refolded protein was then concentrated from 20 mL to a final 500 μL (Vivaspin 10 kDa MWCO; 3000 × g, 15 min at 4 °C) and further purified by SEC (ÄKTA pure system, Superdex-75 Increase 10/300 GL, GE Healthcare). The unfolding and refolding process for GlnBP WT was conducted under the same conditions as described for the GlnBP variants.

The ITC measurements were performed in a MicroCal PEAQ-ITC isothermal titration calorimeter (Malvern Instruments). The prediction ITC software ‘MicroCal PEAQ-ITC Control’ was employed for designing and conducting the experiments. Once the K_d value and the binding stoichiometry (N) were assigned as predefined values, the concentration of both the protein and the titrant (ligand) stock solutions could be calculated by the ‘design-experiment’ function on the software to get an optimal sigmoidal one-site binding curve. GlnBP concentration was assessed using the Nanophotometer (N60 Touch, Implen GmbH) with at least three reading repeats to get accurate determinations of concentration values. For all ITC measurements, the temperature was set at 25 °C with stirring speed at 750 rev/min. The GlnBPs solution (10 μM in PBS buffer pH 7.4, 300 μL) was manually loaded into the sample cell. The titrant (L-glutamine, 100 μM in PBS buffer, pH 7.4) was automatically loaded into the titration syringe and injected in the sample cell with a titration speed of 2 μL every 150 s and a total of 19 injections. As a control experiment, L-glutamine was titrated into the sample cell containing PBS buffer without GlnBPs. All the titration data were analyzed using the MicroCal PEAQ-ITC Analysis Software.

SPR assays were performed on a Biacore T200 (Cytiva) using a CM5 Series S carboxymethyl dextran sensor chip coated with His-antibodies from the Biacore His-capture kit (Cytiva). Briefly, the chips were equilibrated with running buffer until the dextran matrix was swollen. Afterwards, two flow cells of the sensor chip were activated with a 1:1 mixture of N-ethyl-N-(3-dimethylaminopropyl) carbodiimide hydrochloride and N-hydroxysuccinimide according to the standard amine coupling protocol. A final concentration of 50 µg/mL anti-histidine antibody in 10 mM acetate buffer pH 4.5 was loaded onto both flow cells using a contact time of 420 s for gaining a density of approximately 10,000 resonance units (RU) on the surface. By injection of 1 M ethanolamine/HCl pH 8.0, free binding sites of the flow cells were saturated. Preparation of chip surfaces was carried out at a flow rate of 10 µL/min. All experiments were carried out at a constant temperature of 25 °C using PBS buffer (0.01 M phosphate buffer, 2.7 mM KCl, 0.137 M NaCl, pH 7.4) supplemented with 0.05% (v/v) detergent P20 as running buffer.

For interaction analysis, GlnBP-6His (1.5 µM) was captured onto one flow cell using a contact time of 240 s at a constant flow rate of 10 µL/min. This resulted in a capture density of approximately 1200 RU of GlnBP-6His. Eight different concentrations of glutamine (7.8, 15.6, 31.25, 62.5, 125, 250, 500, and 1000 nM) were injected onto both flow cells using an association time of 50 s and a dissociation time of 360 s. The flow rate was kept constant at 30 µL/min. As control, running buffer was injected. The chip was regenerated after each cycle by removing GlnBP-6His completely from the surface using 10 mM glycine pH 1.5 for 60 s at a flow rate of 30 µL/min.

Sensorgrams were recorded using the Biacore T200 Control software 2.0.2. The surface of flow cell 1 was not coated with GlnBP-6His and used to obtain blank sensorgrams for subtraction of the bulk refractive index background with the Biacore T200 Evaluation software 3.1. The referenced sensorgrams were normalized to a baseline of 0. Peaks in the sensorgrams at the beginning and the end of the injection are due to the run-time difference between the flow cells for each chip.

In total, 26 SPR sensorgrams in three sets of measurements were recorded. To correct for remaining drift in the sensorgrams, the initial 60 s of the sensorgrams prior to Gln injection and the last 100 s of the dissociation phase were first fitted with an exponential function, which was subtracted from the sensorgrams. The drift-corrected sensorgrams were fitted to the reaction scheme of Equation 1 based on the differential equations (Schuck and Zhao, 2010; Myszka et al., 1998).dLsurfdt=ktLbulk - Lsurf-dPLdt\begin{document}$$\displaystyle \frac{d\left [\text{L}\right ]_{\text{surf}}}{dt}=k_{t}\left (\left [\text{L}\right ]_{\text{bulk}}\text{\ }-\text{\ }\left [\text{L}\right ]_{\text{surf}}\right)-\frac{d\left [\text{PL}\right ]}{dt}$$\end{document}dPLdt=konLsurfPtot-PL-koffPL\begin{document}$$\displaystyle \frac{d\left [\text{PL}\right ]}{dt}=k_{\text{on}}\left [\mathrm{L}\right ]_{\text{surf}}\left (\left [\mathrm{P}\right ]_{tot}-\left [\mathrm{P}\mathrm{L}\right ]\right)-k_{\text{off}}\left [\mathrm{P}\mathrm{L}\right ]$$\end{document}

where [L]bulk\begin{document}$\left [\text{L}\right ]_{\text{bulk}}$\end{document} = [Gln] and [L]surf\begin{document}$\left [\text{L}\right ]_{\text{surf}}$\end{document} are the free glutamine concentrations in the bulk flow and at the sensor surface, Ptot\begin{document}$\left [\mathrm{P}\right ]_{\text{tot}}$\end{document} is the total concentration of surface-immobilized protein, and PL\begin{document}$\left [\mathrm{P}\mathrm{L}\right ]$\end{document} is the concentration of bound protein complexes. Conversion to the SPR binding response r via [PL]=αr\begin{document}$\left [\mathrm{P}\mathrm{L}\right ]=\alpha \,r$\end{document} and [P]tot=αrmax\begin{document}$\left [\mathrm{P}\right ]_{\text{tot}}=\alpha \, r_{max}$\end{document} leads to fit results for the binding rate constants that are insensitive to the (unknown) conversion factor α, which can be understood from the fact that the quasi-steady-approximation dLsurf/dt≈0\begin{document}$d\left [\mathrm{L}\right ]_{\text{surf}}/dt\approx 0$\end{document} holds for SPR setups (Schuck and Zhao, 2010; Myszka et al., 1998). The association phases of the sensorgrams were fitted with initial conditions [L]surf=0\begin{document}$\left [\mathrm{L}\right ]_{\rm surf}=0$\end{document} and r=0\begin{document}$r=0$\end{document} and fit parameters k_t and rmax\begin{document}$r_{max}$\end{document} at different values of k_on after substitution of k_off by K_dk_on. Prior to these fits with fixed k_on, a remaining small vertical off-set of the sensorgrams was determined as an additional fit parameter in fits with unconstrained, large k_on and subtracted from the sensorgrams. The first 50 s of the dissociation phases were fitted with a single fit parameter k_t for the initial conditions [L]surf=[L]bulk\begin{document}$\left [\mathrm{L}\right ]_{\rm surf}=\left [L\right ]_{\rm bulk}$\end{document} and r=rmax/(1+Kd/[L]bulk)\begin{document}$r=r_{max}/\left (1+K_{d}/\left [\mathrm{L}\right ]_{\rm bulk}\right)$\end{document}, with rmax\begin{document}$r_{max}$\end{document} determined from fits of the association phase of the sensorgram for unconstrained, large k_on. Background-corrected sensorgrams that do not reach binding equilibrium in the association phase (because of small [Gln]), still show marked drifts in binding equilibrium, or do not resolve the initial increase of the binding signal of the association phase (because of large [Gln]) were discarded, which leads to the 13 sensorgrams of Figure 6B,C, Figure 6—figure supplement 1 with fit results for α=1 μM/RU. Fits with, for example, α=1 mM/RU (not shown) lead to practically identical results. All fits were conducted with Mathematica 13 based on the functions ParametricNDSolveValue to obtain numerical solutions of the differential equations and NonlinearModelFit for fitting parameters of these solutions.

The refolded GlnBP(111 C-192C) and GlnBP(59 C-130C) variants were labeled with commercial maleimide derivatives of AF555/AF647 or ATTO 532/ATTO 643 (Gebhardt et al., 2021), and then purified by SEC. The chromatogram of refolded GlnBP(111 C-192C) labeled with AF555/AF647 is shown in Figure 2B, and those of all other variants and dye labeling combinations are displayed in Figure 2—figure supplement 2. First, the His-tagged protein was incubated in 10 mM DTT in PBS buffer for 30 min to reduce all oxidized cysteine residues. Subsequently, the protein was diluted 10 times with PBS buffer and immobilized on a Nickel Sepharose 6 Fast Flow resin (GE Healthcare). The resin was washed extensively with milliQ water followed by PBS buffer pH 7.4. To remove the excess of DTT, the resin was washed with PBS buffer. The protein was left on the resin and incubated overnight at 4 °C with 5–10 times molar dye excess in PBS buffer pH 7.4. Subsequently, the unreacted fluorophores were removed by washing the resin with 6 mL of PBS buffer. Bound proteins were eluted with 800 μL of elution buffer (PBS buffer, pH 7.4, 400 mM Imidazole). The labeled protein was further purified by SEC (ÄKTA pure, Superdex-75 Increase 10/300 GL, GE Healthcare) to eliminate remaining fluorophores and remove other contaminants and soluble aggregates. The selected elution fractions were used without further treatment for smFRET experiments as described below. In general, all experiments were carried out at room temperature using 25–50 pM of double-labeled GlnBP protein in PBS buffer (pH 7.4). Titration experiments were performed by adding specific concentrations of ligand (glutamine) to the buffer.

Single-molecule μsALEX experiments were carried out at room temperature on a custom-built confocal microscope. In short, alternating excitation light (50 μs period) was provided by two diode lasers operating at 532 nm (OBIS 532–100-LS, Coherent, USA) and 640 nm (OBIS 640–100 LX, Coherent, USA). Both lasers were combined by coupling them into a polarization maintaining single-mode fiber (P3-488PM-FC-2, Thorlabs, USA) and subsequently guided into the microscope objective (UplanSApo 60 X/1.20 W, Olympus, Germany) via a dual-edge dichroic mirror (ZT532/640rpc, Chroma, USA). In general, the 532 and 640 nm diode lasers operated at 60 and 25 μW, respectively (measured at the back aperture of the objective), unless stated otherwise. Fluorescence light was collected by the same objective, focused onto a 50 μm pinhole and separated into two spectral channels (donor and acceptor fluorescence) by a dichroic beamsplitter (H643 LPXR, AHF, Germany). Fluorescence emission was collected by two avalanche photodiodes (SPCM-AQRH-64, Excelitas) after additional filtering (donor channel: BrightLine HC 582/75 and acceptor channel: Longpass 647 LP Edge Basic, both from Semrock, USA). The detector outputs were recorded via an NI-Card (PCI-6602, National Instruments, USA) using a custom-written LabView program.

Data analysis for µsALEX was performed using an in-house written software package as previously described (Gouridis et al., 2015). Three relevant photon streams were extracted from the recorded data based on the alternation period, corresponding to donor-based donor emission F(DD), donor-based acceptor emission F(DA), and acceptor-based acceptor emission F(AA). Bursts from single-molecules were identified using published procedures (Nir et al., 2006) based on an all-photon-burst-search algorithm with a threshold of 15, a time window of 500 μs, and a minimum total photon number (F(DD)+D(DA)+F(AA)) of 150, unless stated otherwise in the figure caption.

For each fluorescence burst, the stoichiometries S* and apparent FRET efficiencies E* were calculated and then presented for all bursts yielding a two-dimensional (2D) histogram. Uncorrected apparent FRET efficiency, E*, monitors the proximity between the two fluorophores and is calculated according to E*=F(DA)/(F(DD)+F(DA)). Apparent stoichiometry, S*, is defined as the ratio between the overall fluorescence intensity during the green excitation period over the total fluorescence intensity during both green and red periods and describes the ratio of donor-to-acceptor fluorophores in the sample: S*=(F(DD)+F(DA))/(F(DD)+F(DA)+F(AA)). Collecting the E* and S* values of all detected bursts into a 2D E*-S* histogram yielded subpopulations that can be separated according to their E*- and S*-values. The 2D histograms were fitted using a 2D Gaussian function, yielding the mean apparent FRET efficiency and its SD or width of the distribution. µsALEX assists in sorting single molecules based on their donor/acceptor dye brightness ratio (stoichiometry S*) and uncorrected mean FRET efficiency (apparent FRET E*), which can be related to the mean inter-dye distance (Gebhardt et al., 2021; Lee et al., 2005).

Analysis with mpH²MM was conducted as described previously by the Lerner lab (Harris et al., 2022). In short, the FRET Bursts software (Ingargiola et al., 2016) was used for detecting single-molecule photon bursts using the DCBS (Nir et al., 2006) AND-gate algorithm with a sliding window of m=10 photons searching for instances with an instantaneous photon rate of at least F=6 times the background rate. Afterwards, bursts of such consecutive photons were filtered to have at least 50 photons originating from donor excitation and at least 50 photons originating from acceptor excitation. In the data analysis, the photon stream was then divided into photon streams of different bursts, and a time shift was applied to the acceptor excitation originating photons stream so that their arrival time range overlapped with that of donor excitation originating photon streams. Optimizations were conducted with state models of increasing numbers of states, and the Viterbi algorithm was employed for calculating the integrated complete likelihood (ICL). Optimizing for larger numbers of states ceased once the ICL ceased to decrease between successively larger state models. Optimized models were manually examined, and the optimal state model selected considering the ICL and the reasonableness of the model given prior knowledge based on transition rates and the E* and S* values of the states. After selection of the most-likely state model, the corresponding most-likely state-path determined by the Viterbi algorithm was used to segment bursts into dwells and to classify bursts by which states were present within each burst.

To support the idea that apo and holo states in solution match with that of the crystal structure, we performed a quantitative comparison of inter-dye distances calculated from dye accessible volumes (AV) on structural models of apo and holo protein, and those derived from the experimental smFRET results. For dye AV calculations, we used the FPS method, established by the Seidel lab (Kalinin et al., 2012; Figure 2—figure supplement 1). The experimental data were corrected for setup-dependent parameters according to Hellenkamp et al., 2018; Peter et al., 2022 to obtain accurate FRET values from µsALEX data. Using a Förster distance of 5.2 nm for AF555/AF647, we found good agreement, that is 0.3–0.5 nm deviations (and 1.0 nm in one case) between the calculated and experimentally derived inter-dye distances for both mutants (Figure 2—figure supplement 1).

Solution-based smFRET experiments were performed on a home-built dual-color confocal microscope that combines multiparameter fluorescence detection (MFD) with pulsed interleaved excitation (PIE; Kudryavtsev et al., 2012). MFD-PIE experiments have been described in detail previously (Dahiya et al., 2019). With MFD-PIE, it is possible to extract FRET efficiency, stoichiometry, fluorescence lifetime, and fluorescence anisotropy information from each single-molecule burst. Correction factors including direct acceptor excitation (α), spectral crosstalk (β), and detection correction factor (γ) are also accounted for reporting accurate the FRET efficiency values (Agam et al., 2023). The accurate FRET efficiency (E) can be determined from:E=FGR-αFRR-βFGGFGR-αFRR-βFGG+γFGG\begin{document}$$\displaystyle E=\frac{F_{GR}-\alpha F_{RR}-\beta F_{GG}}{F_{GR}-\alpha F_{RR}-\beta F_{GG}+\gamma F_{GG}}$$\end{document}

where F_GG, F_GR, and F_RR are background-corrected fluorescence signals detected in green/donor (G), red/acceptor (R) after donor excitation and acceptor channels, respectively.

Alternatively, the use of picosecond pulsed lasers and time-correlated single photon counting (TCSPC) electronics enables calculating FRET efficiencies from the quenching of the donor in the presence of acceptor. According to the formula:E=1-τD(A)τD(0)\begin{document}$$\displaystyle E=1-\frac{\tau _{D\left (A\right)}}{\tau _{D\left (0\right)}}$$\end{document}

τD(A)\begin{document}$\tau _{D\left (A\right)}$\end{document} is the fluorescence lifetime of the donor in the presence of acceptor and τD(0)\begin{document}$\tau _{D\left (0\right)}$\end{document} is the fluorescence lifetime of the donor only species. Static species can be observed on the theoretical static FRET line, which is a linear relation between E and τD(A)\begin{document}$\tau _{D\left (A\right)}$\end{document}. Sub-ms conformational dynamics can also be identified and judged by observing the right-shifted populations from the static FRET line.

For the measurements here, 100 pM of GlnBP labeled with ATTO 532 and ATTO 643 was placed on a BSA-passivated LabTek chamber and measured for 2 hours. The sample was excited with 532 and 640 nm pulsed lasers with a repetition rate of 26.6 MHz and laser powers of 45 and 23 µW (measured at the back aperture of the objective), respectively.

Burst-wise FCS analysis is an alternative approach to observe sub-ms conformational dynamics. In this approach, donor (DD) and acceptor (AA) signals detected from single-molecule events are cross-correlated. Thus, fluctuations in the FRET efficiencies appear as an anti-correlated signal in the donor-acceptor fluorescence cross-correlation function. Bursts with sufficient photons detected in both the donor and acceptor channels were selected. A time window of 50 ms was applied around each burst. If another burst was detected within this time window, both were eliminated to ensure correlation functions that are specific to the selected bursts. All the above-mentioned data analysis was done using the PIE analysis with Matlab (PAM) software package (Schrimpf et al., 2018).

Biotin-streptavidin interaction was used to immobilize tagged proteins and labeled DNA on a PEG-functionalized coverslip for single molecule studies. The protein-his-tag and a biotin-NTA chelated with Ni²⁺ were used to mark GlnBP(111 C-192C) labeled with maleimide modified derivatives of ATTO 532/ATTO 643, whilst DNA labeled with Cy3B/ATTO 647 N was directly tagged with a biotin. To prepare a functionalized glass surface, cover slides (1.5 H Marienfeld Superior) were first sonicated in MQ water for 30 min. The slides were rinsed three times with MQ water, sonicated for 30 min in HPLC-grade acetone, rinsed three times with MQ water again. Then, the slides were sonicated with 1 M KOH for 30 min, rinsed three times with MQ water, and dried with a stream of nitrogen air. To remove any organic material left on the surface, the cover slides were plasma-cleaned for 15 min with oxygen. To create a mPEG/biotin–coated surface, the slides were immediately incubated in a 99:1 solution of mPEG3400-silane (abcr, AB111226) and biotin-PEG3400-silane (Laysan Bio Inc) in a Toluene solution overnight at 55 °C. After incubation, the slides were sonicated (10 min in ethanol, 10 min in MQ water), dried under nitrogen stream, and kept under vacuum. Prior to TIRF experiments, each slide was incubated with a 0.2 mg/mL streptavidin in PBS solution for 10 min utilizing Ibidi sticky-slide (18 well) for single molecule studies. PBS buffer pH 7.4 was used to wash away the unbound excess of streptavidin. For GlnBP(111 C-192C) immobilization, 20 nM biotin-NTA (QIAGEN) was charged with 50 nM Ni²⁺ and incubated on the slide for 10 min before rinsing away the unbound excess biotin-NTA and Ni²⁺ with PBS (this step was omitted for the labeled DNA samples). GlnBP(111 C-192C) at 0.8 nM and dsDNA at 0.04 nM were incubated for 5 and 1 min, respectively. For single-molecule data collection, imaging buffer (PBS, pH 7.4) containing 2 mM Trolox was used. For dsDNA, we used PBS buffer in combination with an oxygen scavenging system (pyranose oxidase at 3 U/mL, catalase at final concentration of 90 U/mL, and 40 mM glucose). After that, the chambers were sealed with Silicone Isolators Sheet Material (Grace Bio-labs). All the single-molecule investigations were done at room temperature.

Single-molecule TIRF measurements were conducted on a homebuilt microscope using an Olympus iX71 inverted microscope body. Light from a 532 nm continuous wave laser (532 nm OBIS, Coherent) was transmitted off-axis onto the back-focal plane of a microscope objective (UAPON TIRF 100X1.49 NA, Olympus) via a dual-band dichroic beamsplitter (TIRF Dual Line Beamsplitter zt532/640rpc, AHF Analysetechnik) to generate total internal reflection at the glass-water interface. Fluorescent emission was then split spectrally using a Dual View System (DV2, Photometrics) equipped with a dichroic beamsplitter (zt640rdc, AHF Analysetechnik). The two emission channels were then spectrally filtered using emission filters (582/75 Brightline HC and 731/137 BrightLine HC respectively, both AHF Analysetechnik). Image series were acquired using an EMCCD camera (C9100-13, Hamamatsu) in combination with the µManager (Ingargiola et al., 2016) software. The iSMS (Kalinin et al., 2012) software was used to retrieve and calculate traces of the donor and acceptor fluorescence intensity from consecutive fluorescent images.

Starting point of our atomistic simulations with the AMBER20 software package (Case et al., 2020) and the ff14SB force field parameters (Maier et al., 2015) was the ligand-bound crystal structure with PDB identifier 1WDN. The protein state of the titratable amino acids and including ligand was determined with the software PROPKA3 (Olsson et al., 2011). The protein, with and without ligand, was solvated in explicit TIP3P water in an octahedral simulation box with a minimum distance of 15 Å of protein atoms to the box boundaries at a salt concentration of 150 mM. The two simulation systems with and without ligand were carefully relaxed in nine steps according to the AMBER tutorial ‘Relaxation of explicit water systems’ (see https://ambermd.org/tutorials/). Production simulations starting from the system conformations obtained after relaxation were performed with the AMBER20 software implementation for GPUs (Case et al., 2020; Salomon-Ferrer et al., 2013) with a time step of 4 fs after hydrogen-mass repartitioning (Hopkins et al., 2015). In these simulations, the temperature was kept at 300 K using a Langevin thermostat with a collision frequency of 1 ps^–1 and the pressure was kept at 1 bar with the Berendsen barostat. To observe unbinding events on the microsecond timescale accessible in our simulations, we reduced all interactions between the protein and the ligand by 16% by rescaling the partial charges and ε parameters of the ligand atoms with the commands change charge and changeLJSingleType of the program ParmEd implemented in Amber.