Identifying coupled clusters of allostery participants through chemical shift perturbations

Identifying Residues That Serve as Markers.

We began by acquiring NMR spectra of KcsA in four limiting conditions with respect to ligand binding (Fig. 1B and SI Appendix, Fig. S1). Our goal was to use these spectra to cluster residues into the four groups mentioned above, namely, responsive to K⁺ only, responsive to H⁺ only, responsive to neither, and responsive to both. For KcsA, the apo state can be achieved with 0.1 μM K⁺ at pH 7.5 [referred to as a deeply inactivated state (I*)], the K⁺-bound state with 50 mM K⁺ at pH 7.5 [referred to as the resting or deactivated state (D)], the H⁺-bound state with 0.5 mM K⁺ at pH 3.5 [referred to as the inactivated state (I)], and the doubly ligated K⁺- and H⁺-bound state with 100 mM K⁺ at pH 3.5 [referred to as the activated state (A)]. The conditions for preparing these samples were selected based on our previously measured affinity (K_d) values for K⁺ at pH 7.5 and pH 3.5 (30, 31). Chemical shift assignments for these samples were made based on a dipolar-assisted rotation resonance (DARR) (53) ¹³C-¹³C chemical shift correlation experiment with various mixing times and heteronuclear nitrogen-carbon double–cross-polarization experiments (44) (Materials and Methods). Numerous sites in KcsA had chemical shifts that were invariant to ligand concentrations, confirming that the basic fold of the protein is not affected by K⁺ or pH over a broad range (SI Appendix, Table S1).

KcsA residues that respond to a single ligand are expected to exhibit two chemical shift values in slow exchange, whose peak intensities track the presence or binding of their corresponding ligands. Comparing chemical shifts among this set of spectra in the four sample conditions shows that, indeed, many sites in the protein exhibit two chemical shifts whose intensities depend on a single ligand. Many of these residues are found in the K⁺ binding sites or the pH gate, and were used in prior work to study K⁺ affinity (30, 31) (SI Appendix, Fig. S2). It is notable that residues quite distant from the K⁺ and H⁺ binding site sometimes exhibit two-state behavior (SI Appendix, Fig. 1D and Table S1). Some of the remote reporters responding to K⁺ binding are located behind the selectivity filter, such as E71 and D80; in the outer mouth of channel, such as Y81 and I60; or in the bilayer-embedded helices, such as I38. Interestingly, these residues do not show a clear physical contact network, but are spread out all over the protein. It is worth noting that atoms within the same residue can have a different chemical shift response to torsion or steric interactions, so they do not always reflect the same trends (54) (SI Appendix, Table S1). We discuss details of the assignment of reporters to the two networks more below.

We also observed that a handful of functional groups in key amino acids exhibit more than two chemical shifts, whose intensities reflect the combination of ligand occupancies (Fig. 2A). In some cases, a residue’s side chain can exhibit three distinct shifts, while the backbone atoms exhibit two. Examples include T74 and T75, in which the CG side chain displays three different chemical shifts (Fig. 2B), but the CA and CB side chains only display two chemical shifts each (Fig. 2A). We reason that the side-chain atoms have more flexibility, or are in more direct contact with other residues in the networks of both binding pockets. Multistate behavior can manifest as a change in chemical shift, as illustrated by T74 and T75, or as a change in dynamics, as illustrated by the behavior of residue I100 (44) (Fig. 2B; 1D slices of I100 peaks are shown in SI Appendix, Fig. S3). Allostery has been shown to involve changes in the dynamic behavior of various states as well as in the underlying structure (26, 55). NMR is very sensitive to protein dynamics on time scales from picosecond to seconds (56, 57). For I100, we observed two sets of chemical shifts for the CB-CG2 correlation in the apo, deactivated, and inactivated states; in the activated state (protonated/K⁺ binding state), the peak corresponding to I100 is missing, suggesting that it is broadened due to structural inhomogeneity or that its relaxation times are shorter. In either case, it is reasonable to assume that the structure or dynamics of I100 in the activated state are distinct from those in the other states. For these residues (T74, T75, and I100), the populations of the NMR reporters depend on binding of both ligands such that a distinct form is observed if and only if both ligands are simultaneously bound. From this, we conclude that for residues T74, T75, and I100, the local structure and energetics depend on the combination of the binding status of the two ligands (not simply on one ligand or the other). This observation suggests key roles for these residues in allosteric coupling between the two binding sites. Indeed, the role of I100 in mediating the coupling due to its contacts with F103 has been proposed and examined by various methods (58). Also, the importance of the T75 side chain has been identified by showing that T75A mutant lacks inactivation and shows an inverted coupling (59). The side chains of T74 and T75 are similarly located between the proton and potassium ion binding sites, which further supports their potential role in connecting the two networks (Fig. 2C). In this work (as discussed below), we also examine the role of T74.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Two-state and multistate NMR markers are shown on NMR spectra of samples in various K⁺, H⁺ concentrations. (A) Two typical “two-state” NMR markers for K⁺ binding are illustrated, namely, T74 (blue circle) and T75 (blue dashed circle) backbone CA and CB correlations. (B) Two cases of multistate allostery participants are shown. In the first row, markers I100 CB-CG2 are shown for samples in various conditions and are labeled using a red dashed circle. For samples at low pH and high [K⁺], the peak for I100 is not detected. We surmise that this may be due to dynamics or heterogeneous conformations. In any event, the appearance of the peak depends on the combination of ligands present, and not specifically on one or the other ligand. In the second row, two multiple-state markers are shown, the side-chain CG atoms of T74 (red circle) and T75 (red dashed circle), displaying three distinctive chemical shifts for the four spectra of KcsA at various conditions. (C) Allosteric participants T74, T75, I100, and F103 are highlighted on a crystal structure (Protein Data Bank ID code 1K4C), indicating plausible connections between the two binding sites on the opposite sides of the membrane in KcsA.

Identifying Networks by Chemical Shift Correlation.

The majority of the reporters exhibit two possible NMR peaks over the titration and depend only on one of the ligand binding events. We hypothesized that residues in the same functional unit will experience correlated population changes as ligands are titrated. However, given the strong allosteric coupling, the binding of one ligand will induce binding or release of the other in some regimes, leading to an apparent correlation of the two networks and potential confusion. We clustered the reporters into functional units to identify which ligand each reporter is sensitive to. In other words, for some experimental conditions, a straightforward titration of one variable (e.g., K⁺) can lead to changes in binding at the other (H⁺) binding site. This complicates separation of the two functional units, as residues that are only affected by one ligand appear to respond to the other ligand. For example, E118 and E120 change chemical shifts between the high- and low-K⁺ spectra at neutral pH (44, 47); however, they are known to be the proton binding markers (60). Such correlated binding occurs in a restricted concentration regime. To buttress our assignments of the various reporters distinguishing residues that belong to one reporter network from those in the other network, we include a broad range of conditions, including ones where the binding of the ligands can be separately controlled by concentrations. For example, for KcsA, a K⁺ titration done at pH 3.5, where the pH sensor remains protonated throughout the K⁺ titration, allowed us to uniquely identify K⁺ binding markers. Titrations carried out at pH 7.5 reflect simultaneous changes in both pockets, the direct effects of changing K⁺ occupancy in the selectivity filter, and the secondary effects of changing H⁺ occupancy in the pH sensor; as expected, both K⁺ markers and H⁺ markers change.

Titrations were detected using ¹³C-¹³C 2D spectra of KcsA, in which the populations for various states can be monitored at a large number of sites by integrating the corresponding peaks since the exchange process is slow [consistent with the millisecond-to-second time scale of the inactivation and recovery process as measured by electrophysiology (51, 61) and the exchange rate estimated by NMR as <500 per second (31)]. The assignments of the residues were verified by other 2D or 3D homo- or heteronuclear correlation. Various residues show different magnitudes of chemical shift change in terms of parts per million (0.5–2 ppm) that are useful in this analysis and are enumerated in SI Appendix, Figs. S4 and S5.

As demonstrated by previous NMR studies (32, 62), correlation analysis is a useful tool to evaluate the relationship between two residues. Here, we monitored the population shifts of individual sites as we titrated ligands through the system and then evaluated the Pearson correlation coefficient for residue pairs (as discussed further in Materials and Methods):Conv(X,Y)= E(XY)−E(X)E(Y)/E(X2)−(E(X))2E(Y2)−(E(Y))2, where X, Y represents an array of population percentage of bound state (peaks dominate in high [K⁺] condition) in a series of titration points of the two residues for correlation analysis. From the coefficients, we constructed the covariance matrix of population changes in response to the ligands (Fig. 3 A–D).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Chemical shift correlation analysis reveals networks of residues in KcsA involved in binding. Residues that show two alternative forms are analyzed to discern whether they report on K⁺ vs. H⁺ binding. The covariance of the populations as [K⁺] is titrated, as illustrated using a heat map for pH 7.5 (A) and pH 3.5 (C). To further illustrate, a cross-section at V76 is expanded to show the correlation of V76 populations compared with populations detected for other residues. This analysis was performed at both neutral pH (B, diamonds) and acidic pH (D, rectangles). A network of potassium binding sites is revealed at pH 3.5 (a condition where potassium binding is decoupled from proton binding in the titration) such that they all report on the same binding event with the same apparent affinity. The network is dispersed throughout the protein and includes I60, E71, T74, T75, V76, D80, L81, Y82, and A92. The correlation network at pH 7.5 also includes E118 because potassium binding at this pH is coupled to proton binding in the titration. E118 shows a negative correlation at neutral pH, since the allosteric coupling causes proton release to be correlated with K⁺ binding over a range of solvent conditions. E118 is considered to be a protonation marker for the activation gate. For I38 and E51, we observe very low or no correlation to other residues, such as V76, at both pH values, indicating that the observed chemical shift changes for these residues are likely to be unrelated to binding of these ligands. (Additional correlation plots are shown in SI Appendix, Figs. S4 and S5.)

Our results show that the populations of numerous residues are highly correlated throughout the titration: At pH 7.5, we observed a global network, which includes residues in the K⁺ binding sites, such as V76, T75, and T74; flanking residues, such as I60, E71, D80, L81, and Y82; and proton binding residues E118/E120 (Fig. 3 A and B). As mentioned above, near these conditions, a coupled binding event occurs. By contrast, the decoupled transition at pH 3.5 involves only K⁺ binding. Many residues exhibit correlated shifts and serve as K⁺ binding markers, including residues in the binding site and quite distal residues (Fig. 3C). However, the proton binding residue E118 remains protonated throughout the titration at pH 3.5, as is indicated by the flat slope in its correlation with V76 (Fig. 3D, Upper Right). This flat correlation allows us to filter out this residue as belonging to the proton binding unit, rather than the potassium binding unit, by comparison with the globally coupled network identified at pH 7.5.

We also identified I38 and E51 as residues that show chemical shift changes, but the population changes do not show a strong covariance with other two-state marker sites in response to [K⁺] change (Fig. 3 A and C). E51 is likely a pure proton binding marker, since its chemical shift change is mainly caused by pH change, but not an allosteric marker. For I38, the chemical shift change is more likely to be caused by other environmental factors, which have not been clarified here.

Functional Tests of Putative Allostery Participants.

We next studied the allostery participant sites where the chemical shift serves as a marker for the combination of binding states. Such sites exhibit more than two chemical shifts for four distinct combinations of “gates” or binding states. We propose that they are potentially involved in energetic coupling between the two binding sites. We sought corroborative evidence for the role of putative residues identified through their NMR shift behaviors. We examine the functional behavior of site-specific mutations involving the putative coupling participants and compared with WT. I100 and F103 have previously been suggested as possibly important allosteric coupling residues (41, 58) (Fig. 1C). Mutagenesis, molecular simulation, and our NMR study have provided strong evidence for the participation of F103 and I100 (30, 41).

T74 is another possible important allostery participant in allosteric coupling, as suggested by this CAP analysis (Fig. 3A); it has previously been suspected to interact with F103 and I100 during inactivation (52, 58), but without direct evidence. Its involvement is also difficult to detect simply by comparing crystal structures of various states (SI Appendix, Fig. S7). The observation of “multistate behavior” of the NMR peaks for the T74 CG side chain, and its sensitivity to the combination of binding states, led us to propose that the side chain of T74 is important for propagating allostery. To test this hypothesis, we mutated T74 to a serine to remove the CG methyl group of the threonine and characterized the allosteric coupling by NMR, namely, the K⁺ affinity changes, as a function of pH. As shown in Fig. 4A, the affinity for K⁺ in T74S KcsA decreased only fourfold upon switching from pH 5 to pH 3.5 (red), while for WT KcsA, the K⁺ affinity decreased 250-fold for the same pH change (purple). Thus, the T74S mutation dramatically reduced the pH sensitivity of the K⁺ affinity, indicating that the allosteric coupling network is significantly reduced. This result suggests that channel opening is less coupled to inactivation at the selectivity filter.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Mutation of T74S causes a strong reduction of allosteric coupling, indicating that the side chain is a crucial player for propagating allosteric interaction and C-type inactivation. (A) NMR-detected titration of T74S with respect to K⁺ indicates that the effect of pH on the K⁺ affinity (i.e., coupling strength) is much weaker for the T74S mutant compared with WT. A prior titration study on WT shows that the affinity of K⁺ is around 250-fold weaker at pH 3.5 than that at pH 5.0. (i.e., 4 mM vs.16 μM, the purple symbols in the plot; taken from ref. 30). Titration of T74S in this study indicates that the affinity for K⁺ is only fourfold weaker at pH 3.5 than that at pH 5.0. (i.e., 4 μM vs. 1 μM, the red symbols) The extremely high affinities for K⁺ at both pH values suggest that T74S is not only a coupling mutant but also an affinity mutant similar to E71A (42). (B) Opening probabilities of T74S and WT-KcsA are 0.89 ± 0.12 and 0.014 ± 0.010, respectively. The values are mean ± SD from nine and three experiments for T74S and WT-KcsA, respectively. The lack of inactivation is consonant with our analysis that this mutant has weak allosteric coupling. (C and D) Representative single-channel traces for the T74S mutant and WT-KcsA at a voltage of +100 mV. The much higher open probability of T74S than WT is interpreted to mean that the loss of a methyl group has led to failure of inactivation.

To investigate the impact of the T74S mutation on channel activity, we further analyzed this mutant using single-channel recordings in lipid bilayers. As shown in Fig. 4 B and C, T74S shows long openings and a high open probability. In contrast, WT shows brief openings and a low open probability of 1.4 ± 1.0% (Fig. 4 B and D). The low open probability of WT is consistent with previous electrophysiological data and is a manifestation of activation-coupled inactivation (51, 63). These results confirm that removing the methyl group of T74 in the T74S mutant leads to partially breaking this communication pathway, and to an essentially “inactivation-less” channel variant. It is remarkable that such a simple chemical change could lead to a large effect in thermodynamics and function.