The Open Conformation Satisfies Multiple NMR Experiments on Apo Glutamine-Binding Protein: Overcoming Pitfalls in the Study of Interdomain Dynamics

Abstract

Glutamine-binding protein (GlnBP) from Escherichia coli is a prototypical periplasmic binding protein that has been crystallized in an apo, “open” conformation, with its two domains far apart, and a holo, “closed” form, with proximal domains that engulf the cognate L-Gln ligand. A fundamental question about such large-scale conformational transitions—whether the closed state exists in the absence of ligand—is a matter of controversy in the case of GlnBP. Previously, NMR observations have indicated no evidence of the closed form, whereas experimentally validated computations have suggested a remarkable ~40% population. Here, a paramagnetic NMR strategy designed to specifically detect the putative apo-closed species shows that a major population of the latter is highly improbable. Further, NMR residual dipolar couplings collected under three anisotropic conditions fail to reveal differential domain alignment (which would otherwise signal interdomain dynamics) and establish that the average solution conformation is satisfied by the apo-open crystal structure. Our results indicate that the computational prediction of large-scale interdomain motions is far from trivial and may lead to grossly erroneous conclusions without proper experimental validation.

Graphical Abstract

NMR paramagnetic relaxation enhancements and residual dipolar couplings indicate that apo Glutamine-Binding Protein strongly favors an open conformation in solution. The findings expose serious problems in the computational study of interdomain protein motions.

Dynamical large-scale rearrangements of protein domains are crucial for many biological processes, including enzymatic catalysis, ligand binding and signal transduction.^[1] Part of ATP-binding cassette import systems in Gram-negative bacteria, periplasmic binding proteins (PBPs) are two-domain, 25–60-kDa polypeptides that epitomize such conformational behavior. Not only do they function as primary soluble receptors for a plethora of substrates and are essential for nutrient uptake and chemotaxis, but also serve as excellent models to study the relationship between interdomain dynamics and ligand binding. Numerous crystallographic efforts have shown that, typically, the apo state of a PBP adopts an “open” conformation with a solvent-accessible cleft between the domains, while the holo state exhibits a “closed” conformation, with the ligand occupying the cleft and the domains reoriented to engulf it in a Venus flytrap fashion.^[2] Interpretation of the detailed pictures afforded by crystal structures in terms of macromolecular motions, however, is not straightforward. For example, instances where an apo PBP has been crystallized in different interdomain configurations^[3] may represent either conformers that are in dynamic equilibrium in solution or the result of crystal packing artifacts. Further, even when faithfully capturing solution conformations, the crystallization process may select only highly populated ones, in detriment of scarce, albeit relevant species. These hard-to-disentangle dynamical aspects are therefore preferably tackled in solution, where nuclear magnetic resonance (NMR) spectroscopy plays a prominent role (reviewed elsewhere,^[4] with additional examples in refs [5]).

A fundamental question about PBPs (and other systems where ligand binding is coupled with interdomain rearrangements) is whether the closed conformation is accessible in the absence of ligand. Highly sensitive NMR paramagnetic relaxation enhancements (PREs) from apo maltose-binding protein have revealed a minor (~5% population), semi-closed species in equilibrium with a major open conformer.^[6] While the latter is described by the apo crystal structure, the former is different from the holo crystal form, and represents a potential intermediate that facilitates binding.^[6] By contrast, PREs from apo glutamine-binding protein (GlnBP) have not warranted a closed species, as the apo-open crystal structure satisfied the data well.^[7] This suggests a different binding mechanism for GlnBP, where direct ligand interaction with the open form triggers the closed conformation. Recently, however, a study based on molecular dynamics (MD) simulations and an NMR residual-dipolar-coupling (RDC) experiment proposed that apo GlnBP exists in an equilibrium comprising two major populations of open (~47%) and closed (~40%) forms—that closely resemble the apo-open and holo-closed crystal structures, respectively—along with two distinct, minor species (<10% each).^[5g] Given the differences in sample conditions between the PRE^[7] (100 mM phosphate buffer, pH 7.2, 41 °C) and the MD/RDC work^[5g] (50 mM phosphate buffer, pH 6.8, 25 °C), their combined findings suggest a remarkable behavior for apo GlnBP: a dramatic shift in the equilibrium between the open and closed conformations from a state where the latter is undetectable to one where it represents almost half the population in the system, caused by seemingly small changes in buffer composition and temperature. Below, we thoroughly test this hypothesis via PRE measurements and multiple RDC experiments in solution.

For purely didactical reasons, sample conditions such as the above-specified shall be henceforth indicated by their temperature only. At 25 °C (i.e., the conditions of the previous MD/RDC study^[5g]), the ¹⁵N–HSQC spectrum of apo GlnBP shows a single set of signals (Fig. S1). Therefore, any exchange process, including rearrangement of GlnBP’s small and large domain, must be fast on the chemical shift timescale. Protein conformational/dynamical characterization by PRE owes its versatility to the covalent attachment of a paramagnetic probe to an engineered surface-exposed cysteine. Given the r^–6-distance dependence of the PRE, that associated with a nucleus brought near the probe by interdomain closure will consequently be stronger in a closed conformation than in an open one. Hence the sensitivity of this method, since in a system comprised of species exchanging fast on the timescale of the PRE (ns–μs), the latter is the population-weighted average of the individual PREs and, thus, the contribution from the closed conformation can be detected, even at low population.^{[6, 8]} [This PRE feature has been similarly exploited to study transient intermolecular interactions (e.g., see refs [9]).]

Strong interdomain PREs on backbone ¹H^N atoms from holo GlnBP S51R1 (i.e., the S51C mutant reacted with the nitroxide spin-label MTSL) have already demonstrated the viability of this variant to detect the closed conformation.^{[7, 10]} However, apo GlnBP S51R1 at 25 °C yields weak interdomain ¹H^N-PREs that are readily explained, along with the rest of the data, by the apo-open crystal structure,^[11] as indicated by a low Q-factor^[12] of 16.9% (Fig. 1). [Assuming that the closed species coexists in fast exchange only worsens the fit (Fig. S2A).] Two possibilities solely explain this inability to detect the closed conformation:^[13] (i) it is extremely scarce/inexistent, or (ii) it is scarce and the PRE fast-exchange condition is unfulfilled (i.e., otherwise detectable under fast exchange). That is, a large population of the closed species is readily excluded, as it would otherwise cause an obvious decrease in the corresponding signal intensities in the paramagnetic spectrum (relative to the diamagnetic control), regardless of the exchange rates involved.

Figure 1.

Backbone ¹H^N transverse PREs (H^N-Γ₂) from apo GlnBP S51R1 at 25 °C (50 mM phosphate buffer, pH 6.8, 25 °C). (A) Observed PREs displayed (via the indicated color scheme) on the apo-open crystal structure (PDB ID 1GGG).^[11] Residues unaffected or with unmeasurable PRE due to spectral crowding are colored gray; residues with signals obliterated by the PRE are treated as if displaying the strongest effect. The location of the R1 probe (residue 51) is shown. (B) Agreement between observed PREs and those fit-calculated from the structure (the Q-factor is indicated). (C) The fit in panel B shown as a function of residue number, with observed PREs represented by dots and fit-calculated values by a continuous trace. In panels B and C, experimental error bars are shown, and dots are colored according to the protein region (see panel B). PREs associated with the small domain are of the interdomain kind, as the probe is located on the large domain. All fits of the NMR data in this study were conducted with Xplor-NIH^[18] (see Supporting Information).

Although sensitive to the presence of closed species, the PRE approach is less informative about open forms due to the inherently weak interdomain data involved;^[6] thus, the present study is complemented with RDCs. Measured under anisotropic conditions, RDCs allow the determination of the alignment tensor associated with individual domains, from which relative domain orientations and dynamics (ps–ms timescale) can be inferred.^[14] A good indication that two domains reorient (semi)independently from each other in solution is the ability of one of them to align more strongly within the anisotropic medium.^[15] Thus, the problem becomes one of finding a medium where such behavior is evident. One-bond backbone ¹⁵N–¹H^N RDCs were measured for apo GlnBP under three alignment conditions (Fig. 2A): PEG/hexanol at 25 °C, and Pf1 phage at two different temperatures (37 and 41 °C) and buffer compositions (see Fig. 2‘s caption for details). The datasets therefore span the temperature range established by the previous MD/RDC study^[5g] (and above PRE experiment) and the previous PRE work.^[7] Whereas the induced alignments at 25 and 41 °C are mutually correlated (correlation coefficient, r = 0.95), that at 37 °C is more independent from the rest (r = 0.53, 0.63; Fig. S3).

Figure 2.

One-bond backbone ¹⁵N–¹H^N RDCs (¹D_NH) from apo GlnBP under three alignment conditions. (A) The cumulative distribution of RDCs from all datasets displayed against the apo-open crystal structure (PDB ID 1GGG).^[11] (B–D) Agreement between observed RDCs and those SVD-calculated from the structure under the indicated alignment conditions: PEG/hexanol, 25 °C (PEG/hexanol, 50 mM phosphate buffer, pH 6.8, 25 °C); Pf1, 37 °C (Pf1 phage, 20 mM Tris buffer, pH 7.2, 37 °C); Pf1, 41 °C (Pf1 phage, 100 mM phosphate buffer, pH 7.2, 41 °C). R-factors are indicated. Throughout, only residues in secondary structure elements are considered, and coloring denotes the protein region (see panel B).

For each alignment condition, the RDCs were fit to different regions of GlnBP’s apo-open crystal structure by singular value decomposition^[16] (SVD; Fig. 2B–D and Table 1). In every case, the axial component (D_a) of the alignment tensors respectively estimated from the small and large domain are similar to each other, and to those yielded by fitting both the combined domains and the entire protein (i.e., by additionally including the interdomain linker). An equivalent statement applies to the rhombicity (η) of such tensors. This indicates a similar level of alignment for the domains in all three conditions, a conclusion also obtained via the generalized degree of order (σ),^[17] with σ_small/σ_large values ranging from 1.0 to 1.1, where σ_d is the σ associated to domain d (Table S1).

Table 1.

SVD analysis of one-bond backbone ¹⁵N–¹H^N RDCs from apo GlnBP against its crystal structure ^[a]

Domain [c]	PEG/hexanol, 25 °C [b]				Pf1, 37 °C				Pf1, 41 °C
	N	Da	η	R	N	Da	η	R	N	Da	η	R
S	36	–15.6	0.10	23.0	34	9.3	0.42	17.5	41	–7.2	0.07	22.0
L	52	–15.6	0.14	14.6	48	8.9	0.28	25.5	52	–6.8	0.06	20.6
S+L	88	–16.6	0.06	20.5	82	9.1	0.38	22.8	93	–7.3	0.07	23.1
All	95	–16.5	0.07	20.1	88	9.0	0.40	23.3	99	–7.3	0.08	22.5

^[a]PDB ID 1GGG.

^[b]Alignment conditions (see Fig. 2‘s caption for full details).

^[c]Domain, GlnBP fragment(s) considered in the fit (S, small domain; L, large domain; S+L, small and large domain combined; All, full-length protein). N, number of RDCs. D_a (Hz) and η, axial component and rhombicity of the alignment tensor, respectively. R (%), R-factor.

RDC R-factors^[19] from SVD fits that involve the full-length protein range from 20.1 to 23.3%, indicating agreement between the RDCs and the structure.^[20] In each case, excluding the interdomain linker from the fit has a negligible effect, and allows the direct comparison of the resulting R-factor with the average of those obtained from the individual domains (weighted by the corresponding number of RDCs). Throughout, the average R-factor is only slightly lower (by ~2 percentage points at most) than that offered by the simultaneous two-domain fit, which suggests a high correlation between the relative domain orientations in the crystal and in solution. Such correlation is more aptly expressed via the scalar product between normalized vectors constructed from the elements of the alignment tensors estimated from each domain.^[21] Indeed, scalar product values of 0.95 (25 °C), 0.99 (37 °C) and 0.96 (41 °C) confirm the collinearity of the two tensors and, thus, that the crystallographic interdomain orientation closely represents that of the average structure in solution, under every alignment condition. [Assuming that the closed species coexists, it maximally improves the fit of PEG/hexanol-based RDCs at a 20% population with only a slight R-factor decrease (ratio of 0.9), while more than doubling the PRE Q-factor (Fig. S2).]

The above-presented PRE and RDC data on apo GlnBP agree with the apo-open crystal structure of the protein. Although unable to rule out alternative open species and minor populations (of less than a few percent) of (semi)closed forms (discussed elsewhere^[7]), the PREs categorically exclude a significant proportion of the closed conformation, in concordance with prior results.^[7] In stark contrast, MD simulations from the above-discussed, previous MD/RDC work^[5g] predicted a ~40% population of the closed form. Although such study attempted to validate the computations via ¹⁵N–¹H^N RDCs, the single dataset used was inexplicably sparse, with fewer than half the number of points in any of the three sets considered here—which include the 25-°C dataset, specifically aimed at mimicking that of the MD/RDC work. Moreover, the sparse RDCs yield a considerably worse SVD fit to the apo-open crystal structure (R-factor: 33.6%; calculated here using the published dataset^[5g]) than those reported above (Fig. 2B–D), casting doubt on the quality of the data. Finally, the MD/RDC study included a single-molecule fluorescence resonance energy transfer (FRET) experiment, claimed to additionally confirm the MD results.^[5g] However, the all-important kinetic analysis of the data was omitted, and the FRET efficiency of a purported state was inconsistent across the time traces shown.

The study of large-scale motions in multidomain proteins is essential for understanding their ligand-binding mechanisms. The latter, in turn, are central to engineering endeavors aimed at the development of proteins with novel binding capabilities, an area where both PBPs and computational design play a key role.^[22] However, a systematic evaluation of a number of PBP-based receptors reported to target new ligands has revealed their failure to bind, exposing problems in the computational modeling and subsequent experimental corroboration.^[3e] Similarly, our structural and dynamical characterization of apo GlnBP highlights the need for rigorous validation of theoretical predictions, preferably by experimental techniques that report on the studied conformational transitions as directly as possible, where NMR spectroscopy offers a wealth of options.^[23]