Evidence against the ‘Y-T coupling’ mechanism of activation in the response regulator NtrC

Abstract

The dominant theory on the mechanism of response regulators activation in two-component bacterial signaling systems is the ‘Y-T coupling’ mechanism, wherein the χ₁ rotameric state of a highly conserved aromatic residue correlates with the activation of the protein via structural rearrangements coupled to a conserved tyrosine. In this paper we present evidence that in the receiver domain of the response regulator Nitrogen Regulatory Protein C (NtrC^R) the interconversion of this tyrosine (Y101) between its rotameric states is actually faster than the rate of inactive/active conversion and is not correlated to the activation process. Data gathered from NMR relaxation dispersion experiments show that a subset of residues surrounding the conserved tyrosine sense a process that is occurring at a faster rate than the inactive/active conformational transition. We show that this process is related to χ₁ rotamer exchange of Y101 and that mutation of this aromatic residue to a leucine eliminated this second faster process without affecting activation. Computational simulations of NtrC^R in its active conformation further demonstrate that the rotameric state of Y101 is uncorrelated with the global conformational transition during activation. Moreover, the tyrosine does not appear to be involved in the stabilization of the active form upon phosphorylation, and is not essential in propagating the signal downstream for ATPase activity of the central domain. Our data provides experimental evidence against the generally accepted ‘Y-T coupling’ mechanism of activation in NtrC^R.

Introduction

Two-component systems are the predominant signaling systems in bacteria, allowing them to react to a wide variety of changes in their cellular environment. A basic two-component system consists of a histidine kinase whose conserved C-terminal transmitter domain becomes phosphorylated in response to a signal in the variable sensor domain. The histidine kinase’s cognate response regulator can then catalyze the phosphoryl-transfer to its own conserved aspartate residue. In the response regulator Nitrogen Regulatory Protein C (NtrC), which activates transcription of a number of genes under conditions of limited nitrogen supply, the receiver domain is in a pre-existing equilibrium between the active and inactive substates in the wildtype protein.¹ When its associated histidine kinase, NtrB, is phosphorylated due to low intracellular nitrogen concentrations, the subsequent transfer of the phosphate to the receiver domain of NtrC causes a shift in this pre-existing equilibrium far toward the active state through active state stabilization.² The structural change in the region of α-helix 3 through β-strand 5 (‘3445 face’) in the receiver domain (NtrC^R) modifies its interaction with the central domain of NtrC³, leading to oligomerization of the full length protein⁴, ATPase activity⁵, and the activation of the σ⁵⁴ promoter through a DNA looping mechanism.⁶

The activation mechanism of response regulators has gained much attention due to its central role in bacterial signaling and its early recognition as a classical system to study allostery in single domain proteins.⁷ Because our new results presented here rule out the popular ‘Y-T coupling’ mechanism⁸, we briefly review previous publications on this topic. Crystal structures of the homologous response regulator CheY led researchers to propose the ‘Y-T coupling’ mechanism of activation, which is reminiscent of an induced fit view of allostery.⁸ The Y and T in question are Y101 and T82 (NtrC^R numbering) which are highly conserved as either a tyrosine or phenylalanine, and as a threonine or serine, respectively, in response regulators.⁹ A crystal structure of the BeF₃⁻ phosphate analog activated receiver domain of CheY showed that the position of the threonine was coupled to the χ₁ rotameric state of the tyrosine when compared to the unphosphorylated inactive protein.⁸ In both a low resolution¹⁰ and a high resolution¹¹ structure of CheY, in the inactive state the tyrosine was found to be occupying both the g+ and t rotameric states while in activated CheY the threonine moved to coordinate the phosphate group, leaving a gap in the side chain packing. Tyrosine was then thought to fill this gap by going exclusively into the t state and becoming buried in the interior of the protein. This mechanism was also supported by the crystal structures of a series of mutations of the threonine and tyrosine, in which the rotameric state of the aromatic ring corresponded to whether or not the mutants were able to function properly during chemotaxis assays in E coli.¹² In 2006, Dyer and Dahlquist solved a crystal structure of unphosphorylated CheY with a fragment of its target protein, FliM.¹³ They found that although the tyrosine was buried in the interior of the protein, the threonine, as well as the loop between β-strand 4 and α-helix 4 (β4/a4 loop), appeared to be only midway to the activated state. Analysis of all known structures of CheY led to the proposal that before phosphorylation the tyrosine was free to move independently from the threonine and the b4/a4 loop region, but upon phosphorylation, the threonine acted as a gate to the loop motion leading to a correlated movement of the tyrosine rotating to the buried t state. The authors dubbed this mode of allosteric activation ‘T-loop-Y coupling’.^{13; 14} Importantly, activation coupled to the rotameric state of the tyrosine has been proposed to be a general mechanism of response regulator activation based on the tyrosine having distinct states in the inactive and active structures of a number of different response regulators.^{15; 16; 17; 18; 19}

Besides these experimental studies, many computational studies have been carried out on various receiver domains of response regulators with controversial results. Although this tyrosine was seen to be undergoing free rotameric rotation in the inactive state in molecular dynamics (MD) simulations in CheY, the motion of the β4/a4 loop and the tyrosine were reported to be correlated in simulations of the active conformation.²⁰ A similar result was found for FixJ where the β4/α4 loop and phenylalanine appeared to be correlated but the threonine movement was not²¹. In a more recent paper, the tyrosine was shown to be coupled to both the threonine and the β4/a4 loop using a statistical mechanics model of protein allostery in CheY.²² In NtrC^R, elastic network modeling found that motions encompassing the aspartate that becomes phosphorylated, as well as the threonine and the tyrosine, were all correlated.²³ Pandini et al also found support for a pre-organized network of allosteric connections between these residues and the functional site of phosphorylation.²⁴ Alternatively, transition path sampling in combination with potential mean force calculations found that the energy barrier of tyrosine/phenylalanine rotamer exchange was lower than the barrier involved in the inactive/active transition, making the rotameric state of the conserved aromatic residue kinetically independent from the functional conformational transition, and leading them to theorize a role for the tyrosine in the thermodynamic stabilization of the active substate^{25; 26}.

Early NMR experiments to probe protein dynamics gave a hint that the tyrosine may not be directly coupled to the motion of the inactive/active transition. Dynamics on the microsecond to millisecond timescale were seen in residues surrounding Y101 in the phosphorylated form of NtrC^R however, the applied model-free analysis did not allow us to quantitatively characterize conformational exchange processes.¹ Microsecond to millisecond motion has also been seen in phosphorylated CheY, with the majority of it localized to the area encompassing the β4/α4 loop and the tyrosine.²⁷ Despite the plethora of contradicting computational and experimental studies involving the role of the highly conserved tyrosine in the mechanism of activation of response regulators, ‘Y-T coupling’ is the dominant description of activation for these systems.^{14; 17; 18; 28}

Here we use NMR relaxation dispersion experiments in combination with molecular dynamics simulations to provide evidence that Y101 in NtrC^R moves independently, and more quickly than the inactive/active conformational transition both before, and after, the addition of the BeF₃⁻ phosphate analog. By experimentally characterizing a Y101L mutant, we demonstrate directly, that Y101 is not needed for the inactive/active transition and that it does not thermodynamically stabilize the active state structure. This data challenges the prevailing description of the activation of allosteric response regulators.

Results and Discussion

Y101 rotameric state is kinetically independent from the inactive to active transition

Using ¹⁵N NMR CPMG (Carr-Purcell-Meiboom-Gill) relaxation dispersion experiments originally described by Loria et al.²⁹ and Vallurupalli et al.³⁰, in combination with an independently determined exchange-free transverse relaxation rate³¹, we determined the existence of a second dynamic process in NtrC^R, in addition to the previously described inactive/active transition.² For a resonance undergoing exchange, the transverse relaxation rate (R₂^eff) is equal to the intrinsic rate of relaxation (R₂₀) plus an additional rate of relaxation due to exchange (R_ex). The latter can be suppressed by an external magnetic field, a phenomenon that has extensively been used to quantify conformational exchange processes in the millisecond to high microsecond time regime in CPMG experiments^{29; 32;33}. Typical CPMG relaxation dispersion profiles are shown in Fig. 1 A-C, where increasing CPMG frequencies (ν_CPMG) are applied to suppress the R_ex component of R₂^eff with full suppression at the highest field strength of 1000 Hz that should correspond to the exchange-free transverse relaxation rate.²⁹ These data are routinely fit to an equation describing a two-site exchange process to obtain the corresponding rates. However, we discovered that for a number of residues, the seeming exchange-free relaxation rate determined from the CPMG experiments (R₂β_1000Hz) is actually not the true R₂₀β by measuring the latter via a combination of three TROSY based NMR relaxation experiments as described in detail elsewhere.³¹ These values are significantly lower (shown as values for ν_CPMG→∞) than the one predicted from the CPMG curves, revealing a second dynamic process.

Figure 1. Y101 dynamics is kinetically independent from the inactive to active transition.

(A-C) ¹⁵N-CPMG relaxation dispersion profiles and R₂₀β values for residues undergoing exchange between the inactive and active states (k_ex) are compared to the true exchange-free transverse relaxation rates determined independently (R₂₀β)³¹ for A) S85D, B) Y101F, and C) BeF₃⁻ -activated NtrC^R. The CPMG curves refer to residues 78 (cyan), 79 (yellow), 82 (magenta), 88 (orange), 90 (light brown), 100 (purple), 101 (red), 102 (blue), and the independently determined (R₂₀β) are shown as R₂^eff at ν_CPMG→∞. The difference between R₂₀β and R₂β_1000Hz is plotted for all amides in panels (D-F), showing that many residues sense a second fast exchange process that cannot be suppressed by the applied CPMG field strength. G) The residues with exchange from both processes, the inactive/active interconversion and this second faster process (orange) are clustered around Y101 (blue) in both the inactive (pdbID 1DC7) and active (pdbID 1DC8) state structures.

We note that this “hidden” second process could not be unambiguously identified for unphosphorylated wildtype NtrC^R (WT) because its rate of inactive/active interconversion is ~14,000 s⁻¹; too fast to be suppressed by the maximum CPMG field strength of 1000 Hz.² However, for unphosphorylated S85D, Y101F along with WT activated by the phosphate mimic BeF₃⁻, the inactive/active transition is slow enough² to separate the dynamics of the activation step from the faster second process.

By comparing R₂β_1000Hz and R₂₀β we were able to probe on a per-residue basis both dynamic processes. The values for R₂β_1000Hz and R₂₀β calculated for most amides agree well with each other, but a small percentage of residues in S85D, Y101F, and BeF₃⁻ -activated NtrC^R have R₂β_1000Hz values significantly higher than the corresponding R₂₀β (Fig. 1D-F). This provides evidence that in NtrC^R a second faster exchange process is sensed in a subset of residues, which is not fully suppressed in the CPMG experiment, and which does not have any perceptible effect on the CPMG dispersion curves itself. Interestingly, the amino acids with exchange from both processes (orange, Fig. 2G) cluster around Y101 (blue) suggesting that this second faster process is linked to Y101 dynamics. Furthermore, it implies that this movement is unrelated to the inactive/active interconversion, in sharp contrast to the dominant ‘Y-T coupling mechanism’.

Figure 2. Y101L mutant corroborates that the faster dynamic process is due to rotamer exchange.

A) ¹⁵N-CPMG relaxation dispersion profiles for the Y101L NtrC^R mutant. Residues whose dispersion was fit globally to the fast form of the Carver-Richards equation in order to determine the rate of exchange (k_ex) between the inactive and active forms of the protein are shown on the structure (1DC7) in red. Yellow denotes residues whose peaks are severely exchanged broadened, residues that are flat/non-exchanging are colored in blue, and unassigned/proline/overlapped residues are in gray. Dispersion data is color coded as follows: 12 (slate), 13 (pink), 16 (rose), 30 (purple), 40 (dark grey), 41 (light orange), 54 (maroon), 60 (yellow), 68 (light pink), 78 (cyan), 87 (yellow), 88 (orange), 106 (green), 107 (dark green), 108 (olive), 110 (light yellow), 114 (tan), 115 (light green). Importantly, exchange contributions to R₂^eff are mostly suppressed by the applied CPMG field strength, illustrated by the difference between the exchange free transverse relaxation rates determined independently (R₂₀β at ν_CPMG→∞) and from ¹⁵N-CPMG dispersion experiments (R₂β_1000Hz) (B). C) The Y101L mutation (blue) shifts the inactive/active state equilibrium toward the inactive state compared to WT (red) as highlighted in the ¹⁵N-HSQC overlay for residue 88. The 50% active D86N/A89T mutation is shown in orange for comparison.

Y101L mutant verifies that the faster dynamic process is due to rotamer exchange

In order to test this hypothesis, we mutated Y101 to a leucine. This might be considered a dramatic substitution, as the residue in position 101 is typically conserved as either a tyrosine or a phenylalanine in the NtrC family, with only less than 2% of occurrences as leucines and high conservation across response regulators (Figure S4A). The Y101L NtrC^R mutant could be expressed and purified with no apparent differences in expression levels or stability compared to WT. A backbone amide ¹⁵N-CPMG dispersion experiment was carried out on the Y101L mutant and the residues undergoing exchange were fit globally with a rate of exchange equal to 2,200 ± 300 s⁻¹ (Figure 2A), similar to the rate constants of Y101F.²

Strikingly, the second fast process is absent in Y101L as determined by the fact that the R₂₀β’s and R₂β_1000Hz values from the CPMG experiments are identical (Fig. 2 A, B). Only one residue has a statistically significant difference in value of R₂₀β, residue S13. This serine is located in a loop near the phosphorylation site, next to D10 and D11 that help in the coordination of a Mg²⁺ ion upon phosphate binding. Since S13 is far from position 101, it is reasonable to hypothesize that S13 is not sensing the second fast process in WT, but possibly the binding and release of positively charged sodium ions in the buffer.

Two additional observations from the NMR spectra of the Y101L mutant are worth mentioning. First, the equilibrium of the two conformational substates is shifted more toward the inactive state (Fig. 2C), as revealed by the peak position relative to WT and an activating mutant D86N/A89T.¹ Second, residues with dispersion were not only found on the ‘3445 face’, as in previously studied WT, mutants, and activated-NtrC^R, but were also observed along α-helix 5. By removing the aromatic sidechain from the residue at position 101 in NtrC^R, α-helix 5 now appears to be able to sense the dynamics of the activation transition. We note that chemical shift differences between states do not mean that these residues move, but rather that this nucleus has different electronic environments in the exchanging states. A change in the allosteric network upon mutation of the conserved aromatic group has been observed previously in the receiver domain of RegA, with new dynamics measured in α-helix 1 with a constitutively active tryptophan mutation.³⁴

In summary, NMR experiments on Y101L and BeF₃⁻-activated WT NtrC^R provide confirmation that the interconversion between rotameric states of Y101 is the cause of the discrepancy in R₂₀β and R₂β_1000Hz values in S85D, Y101F, and BeF₃⁻ -activated NtrC^R. The rate of the tyrosine flip is faster than the measured rate of the inactive/active conformational change, even in the activated protein, signifying that ‘Y-T coupling’ is not rate-limiting in the activation mechanism of NtrC^R.

MD simulations identify rotameric exchange as independent from activation dynamics

To gain more insight at an atomistic level on the nature of the fast motion involving Y101, its correlation with T82, and the connection of this motion with the global active to inactive state interconversion, we performed a long unbiased MD simulation of the apo active conformation of NtrC^R using the supercomputer Anton. During the 71 μs long simulation we observe that the side chain of tyrosine 101 is not locked in a single conformation, but alternatively interconverts among the three accessible rotameric states (Fig. 3A). The three rotameric states have relative populations of approximately 0.68, 0.28, and 0.04, for the t, g+ and g− states, respectively, and the typical residence time in the most populated one is on the order of 1 μs (see Fig. S1A). The state with the largest population, state t, not surprisingly corresponds to the χ₁ angle that is most often associated with the active state X-ray structures of response regulators (Fig. 3D, green).

Figure 3. Within the active state several rotameric states are accessible to Y101.

A) Time series of the χ₁ angle of Y101 measured along a 71 μs long molecular dynamics simulation of the nonphosphorylated receiver domain of NtrC^R in the active conformation. The region of helix α4–strand β4 is shown above for representative structures of the three Y101 rotameric regions sampled. B) Time series of the χ₁ angle of T82 measured along the same MD simulation. C) Time series of a pseudo dihedral angle reporting on the rotation of the α4–strand β4 around its axis. As a reference, the value of the same order parameter for the inactive conformation of NtrC^R is shown as a thick grey line demonstrating that NtrC^R does not visit the inactive conformation during the 71 μs MD simulation. D) Comparison between the models for the inactive and active structures of CheY with the different χ₁ angles of Y106 and the corresponding T87 orientation (pdbID 1FFG and 1FQW shown in blue and green, respectively) and the structures for the three rotameric states of Y101 with the corresponding T87 orientations visited during the simulation of the active conformation of NtrC^R.

MD simulations can directly address the question of correlated motions that cannot be directly extracted from ensemble-averaged experiments. The simulation shows no strong correlation between the motions of Y101 and T82. The side-chain of T82 remains in the same rotameric state for almost the entire length of the simulation, independent of the position of Y101 (Fig 3B). Interestingly, it is only when Y101 is completely moved away from the active site, as in the minor state g−, that T82 has the freedom to explore all three side-chain rotamers with sizable probability (Fig. S1B). The key result from the simulations is that despite the different conformations attained by Y101 and T82, the rotation of α–helix 4 around its axis, which reports on the crucial change in the solvent exposure of L87, A90, and V91 and is important in downstream signaling, remains in the active conformation throughout the simulation (Fig 3C). This long unbiased MD simulation delivered crucial information about the atomistic details of the Y101 dynamics as well as information about the question of correlated motions, neither of which could be addressed directly by experiments.

Using a combination of computation and experimental data we have shown that the fast timescale motion of Y101 is a rotamer interconversion that is uncorrelated with the global conformational transition that the ‘3445’ signaling face experiences during the activation/inactivation process in NtrC^R, and not correlated to the T82 dynamics, ruling out a Y-T activation mechanism for NtrC as had been proposed in earlier studies.¹⁹ In this context it is important to point out that the structural description of active and inactive conformers in NtrC^R is quite different from the one typically used when discussing receiver domains of other response regulators. In CheY the structural variation between the active and inactive conformers is more localized than in NtrC and can be described by changes in the positions of a handful of residues, including T82, Y101 and the loop between β-strand 4 and α-helix 4, whereas the conformation attained by α–helix 4 is nearly identical in the two conformers.³⁵ In NtrC^R, a more global conformational change occurs and this global change is essential for signal propagation. For this reason, the rotameric state of Y101 may be less important in the activation mechanism and downstream signaling of NtrC^R as compared to other response regulators, such as CheY.

Comparison between NMR experiments and MD simulations

The MD simulations on activated apo NtrC^R suggest that the second faster process detected by the NMR relaxation experiments may be a rotamer exchange of Y101, in agreement with the NMR data on Y101L. Three bond J-coupling values deliver valuable information about the χ1 rotamer angles of aromatic residues by NMR.³⁶ To test our hypothesis, we determined the ³J_NCγ value for Y101 in BeF₃⁻-activated WT NtrC^R to be 1.6 +/ 0.2 Hz, where gauche χ1 angles cluster around 0.5 Hz and trans angles cluster around 2.4 Hz. Aromatic residues that sample more than one rotamer in the fast NMR time regime will have an intermediate value. The ³J_NCγ for Y101 in the activated state is consistent with the tyrosine sampling multiple rotamers with a higher population of the trans state as seen in the MD simulations (Fig. 3A).

From the NMR relaxation data a rough estimate of the rate constant of about 30,000 ± 12,000 process was determined for BeF₃⁻-activated NtrC^R (see supplemental material). We want to note that it is impossible to separate populations from delta omegas in the fast exchange limit. In principle, one could try to estimate the chemical shift difference from the MD simulations and use this information as input for fitting the NMR data. While qualitatively chemical shift perturbations estimated from the MD simulations using Sparta+³⁷ are in the same regions as detected by NMR (Fig. S2), this approach does not yield reliable quantitative results (Fig. S3), reaching the existing current limits in our methods. In the MD simulations three different rotameric states are being sampled with different lifetimes. First, even with a 71 μs simulation, insufficient sampling hampers accurate estimations of populations and rate constants. Second, a multi-state conformational exchange on this timescale cannot be quantitatively dissected from the NMR experiments. Third, the slower and the faster processes measured by NMR superimpose for many residues, and both processes are not necessarily uncorrelated. For these reasons we prefer not to over interpret our data and only calculated a rough estimate of the second process.

Aromatic rotamer dynamics in proteins has been extensively studied and continues to be a topic of interest. MD and fluorescence studies on rotamer exchange on small peptides have found rates of interconversion for aromatic residues on the nanosecond timescale³⁸, but the variability in packing of aromatic residues in proteins can lead to a large range of ring flipping frequencies, spanning from the minutes to the subnanosecond timescale, even within the same protein.^{39; 40; 41;42} In our MD simulation, F99 does not rotate, whereas Y122 rotates in sub-microseconds within NtrC^R.

Mutating the conserved aromatic does not affect activation and NtrC’s downstream ATPase activity

As previously mentioned, the amino acid corresponding to position 101 in NtrC^R is predominantly an aromatic amino acid in all receiver domains throughout the whole two-component system superfamily. Specifically, for NtrC, that amino acid is a tyrosine or phenylalanine in approximately 90% of the sequences classified as NtrC in the prokaryotic two-component systems database,⁴³ with only few occurrence of other amino acids (Fig S4A). Other subfamilies, such as CheY, CheV and CheB, show a slightly higher variation (Fig S4B-D), but in all analyzed response regulators families, the residue has an aromatic side chain in about 85% of the sequences (Fig S4F). The high level of conservation, together with the available structural information of active and inactive conformers, led researchers to identify the tyrosine as crucial trigger in the activation mechanism of the receiver domain of response regulators.⁸ However we have shown here that the conformational changes that Y101 undergoes are not related to the global conformational rearrangement hallmark of the activation of NtrC.

It is instead possible that, as suggested by other studies,^{25; 26} the aromatic residue in position 101 is needed thermodynamically in the stabilization of the active state once the protein becomes phosphorylated. To test this we activated Y101L NtrC^R with BeF₃⁻, a ground state phosphate mimic that has been shown to activate NtrC^R to the same extent as phosphate.^{19; 44} An [¹H¹⁵N]-HSQC overlay of BeF₃⁻ -Y101L with BeF₃⁻ -WT protein show similar spectra, with the peak positions confirming that the activated Y101L mutant shifts its population toward the active state similarly to WT (Fig. 4B). BeF₃⁻ -activated Y101L NtrC^R also shows an exchange rate comparable to BeF₃⁻-activated NtrC^R (1,500 ± 600 s⁻¹ and 1,950 ± 290 s⁻¹ respectively) (Fig. 4A). These results imply that stabilization of the active state is also not dependent on the conserved tyrosine residue.

Figure 4. Mutating the conserved aromatic does not prohibit activation of the receiver domain or NtrC’s downstream ATPase activity.

A) ¹⁵N-CPMG NMR relaxation dispersion profiles for the BeF₃⁻ -activated Y101L NtrC^R mutant fit to a single exchange process of inactive/active interconversion. Residues whose dispersion was fit globally to determine the rate of exchange (k_ex) between the inactive and active forms of the protein are shown on the structure (pdbID 1DC8) in red. Yellow denotes residues whose peaks are severely exchanged broadened, residues that are flat/non-exchanging are colored in blue, and unassigned/proline/overlapped residues are in gray. Dispersion data is color coded as follows: 6 (ivory), 10 (burnt sienna), 12 (slate), 53 (tan), 64 (brown), 78 (cyan), 81 (green), 82 (magenta), 96 (purple). B) ¹⁵N-HSQC overlay of WT, Y101L, BeF₃⁻ -activated WT, and BeF₃⁻ -activated Y101L NtrC^R highlighting that the Y101L mutant has the characteristic peak shifts associated with the active state upon the addition of BeF₃⁻. C) Full length NtrC ATPase activity assays comparing the net activity of BeF₃⁻ -activated WT (dark grey) and Y101L (light grey).

Since we have shown that Y101 is not necessary for the kinetics or thermodynamics of the functional conformational change in the receiver domain of NtrC^R, the next logical possibility to investigate is a potential crucial role of this conserved aromatic residue in the propagation of this structural change to the next step in the signaling cascade. Stabilization of the active conformation upon phosphorylation exposes a group of hydrophobic residues (L87, A90, and V91) on the ‘3445 face’ which have been suggested to be the contact sites between the receiver domain and the central domain of NtrC.¹⁹ This central domain is responsible for both the oligomerization and ATPase activity of the protein, leading to the activation of the σ⁵⁴-holo enzyme transcription factor. To assess the role of Y101 in the signal transduction between receiver and central domain, we measured and compared the ATPase activity of BeF₃⁻ -activated full-length WT and Y101L NtrC using a colorimetric assay that measures the amount of inorganic phosphate released due to ATP hydrolysis (Fig. 4C). The results show that BeF₃⁻ -activated Y101L NtrC has only a small decrease in the rate of ATP hydrolysis compared to WT indicating that an aromatic residue at position 101 is not essential for transducing the signal to the downstream central domain of the protein.

Similar mutational studies on the activity of homologous response regulators have been carried out previously on CheY and OmpR. A study by Zhu et al.⁴⁵ on CheY looked at the effect of mutations on the ability of CheY in E. coli to direct flagella rotation by monitoring swimming and tumbling during chemotaxis. They found that replacement of the tyrosine by a tryptophan led to a hyperactive phenotype. The mutation to a phenylalanine, conversely, created a phenotype that had very low activity, while the leucine, valine, isoleucine, and glycine mutations resulted in no measureable activity at all. In an X-ray crystal structure of a constitutively active mutant, D13K/Y106W, bound to a peptide of its downstream interaction partner, FliM, CheY was found to be bound to FliM and in the active conformation in two of the four molecules in the asymmetric unit, while the other two molecules were unbound and in the inactive conformation.⁴⁶ In the Y106W single mutant structure, however, CheY was found to be in the active form in the absence of peptide.¹² Based on these results, the authors suggested that there might be a low energetic barrier between the inactive and active conformations in CheY, which includes the conserved aromatic residue. This barrier may increase once the protein is bound to its downstream target, FliM.³⁸ In the OmpR system, which is involved in osmotic regulation, a mutation of the conserved tyrosine to a cysteine resulted in a constitutively active protein as monitored by gene expression.⁴⁷ This mutation also caused an increase in the ability of OmpR to bind to its cognate DNA in vitro. Interestingly, in both of these cases, the rate of phosphorylation and dephosphorylation of the mutant proteins remained the same as for WT suggesting that the difference in activity may not be caused by the receiver domain activation but by the domain’s ability to transmit the signal downstream.

Conclusion

The activation mechanism of NtrC^R appears to be quite different from what has been previously proposed.^{8; 13} Our data, together with others’^{1; 25; 26; 27}, shows in fact that the ‘Y-T coupling’ model of activation, inferred mainly from structural differences in the highly conserved Y101 between the active and inactive conformations of various response regulators, is not valid for NtrC. The motion of Y101 is uncorrelated to the global conformational rearrangement and the aromatic side-chain is not needed for the stabilization of the activated state by BeF₃⁻ or for activating the central domain ATPase. These observations highlight the fact that by looking beyond static structures and focusing further on ensembles that correspond to the inherent dynamics in proteins, more accurate predictions about whether, and how, specific residues are implicated in function may be obtained. Although harder to come by, information about structures and kinetics of lowly populated states along the transition have the possibility of providing a wealth of information that is not accessible by looking only at the more highly populated ground states that are typically captured in structural biology.

NMR dynamics experiments are becoming widely used to ascertain information about the interplay between protein motion and function.^{48; 49; 50} Our results illustrate that interpreting dynamic data can be complicated when more then one dynamic process is present, which is a common trait of biological systems. In some cases three-state models can be quantified by combining multiple CPMG dispersion experiments⁵¹ or by doing the experiments at multiple field strengths and temperatures^{52; 53; 54; 55; 56}, but in those cases the R₂^eff is fully suppressed by the maximum field strength. By measuring the exchange free transverse relaxation, in addition to CPMG dispersion for NtrC^R, we were able to separate the inactive to active conformational transition from an independent tyrosine rotation for residues where R₂^eff was not fully suppressed. In addition, we used the single molecule characteristic of molecular dynamics simulations to address the question of correlated motions. Recently, CPMG NMR dispersion experiments on non-phosphorylated CheY were fit to a strict two-state model, where Y101 was fit to a slightly faster rate compared to other residues.⁵⁷ The data were interpreted with a model of inactive to active switching in a non-concerted sequential fashion with a possible preferred sequential order. In light of our results presented here, we wonder if the CPMG data on CheY reflect a similar dynamic mechanism where the tyrosine rotation is a second independent process from the activation, as we have described on NtrC. Separating these different motions, by acquiring additional dynamics data, might give more insight into the possible multiple motions that CheY is experiencing and better assign a biological role to these processes.

This new information on the activation mechanism of response regulators stimulates further questions regarding the reason why this aromatic residue is so highly conserved in this family of proteins. As already shown in CheY,⁴⁵ and also seen in our work on NtrC, replacing the tyrosine with a non-aromatic residue does not impair the ability of the receiver domain to be phosphorylated. Moreover, we have shown that the receiver domain of NtrC lacking Y101 is able to transfer the signal to the next downstream partner, the central ATPase domain. Based on these observations, we speculate that the tyrosine may play a role in interactions with further downstream partners, such as the σ⁵⁴-holoenezyme transcription factor. This hypothesis is in line with activity data in both CheY and OmpR, where mutating the tyrosine led to a change in functionality at the cellular level. It is also possible that the aromatic residue is important to interactions with other regulatory domains in some two-component systems, such as the P2 domain on CheY’s associated histidine kinase, CheA⁵⁸ or with associated phosphatases.⁵⁹ Our results underscore the importance of determining the biological roles of specific dynamic processes. This new perspective on the activation mechanism and signal propagation in NtrC^R could provide new insights into two-component signaling in bacteria and regulation by allosteric proteins in general.

Methods

NMR sample preparation

Unlabeled and uniformly ¹⁵N-labeled NtrC^R wild-type, mutant, and BeF₃⁻ activated forms were prepared as previously described. ^{1; 19} S85D and Y101F NtrC^R NMR samples were 0.75 mM in 50 mM Na₃PO₄ buffer, pH 6.75 with 10% D₂O. Y101L, BeF₃⁻ Y101L, and BeF₃⁻ WT NtrC^R samples were 0.3 mM to avoid transient aggregation.

NMR experiments

TROSY ¹⁵N relaxation dispersion experiments were acquired on Varian Inova 600 MHz spectrometer at 298K. Constant time T₂ delays between 60 and 70 ms were used in all TROSY CPMG dispersion experiments,^{29; 60; 61; 62} which roughly yielded 55% of residual signal intensity, and CPMG field strengths between 28 and 1000 Hz were collected. Data was processed using NMRPipe⁶³ and the change in R₂^eff with increasing frequency of CPMG refocusing pulses were fit globally for each mutant to the fast form of the Carver-Richards equation (R₂^eff (ν_CPMG) = R₂₀β(ν_CPMG→∞) + (p_Ip_AΔω²_IA/k_ex)[1-(4ν_CPMG/k_ex)tanh(k_ex/4ν_CPMG)]) in Matlab. The variation of R_ex with an increasing applied radio-frequency field (ν_CPMG) permits the CPMG dispersion experiment to quantitatively determine kinetics (k_ex) for a two-state exchange process^{29; 61} even for highly skewed populations.⁶⁰ Noise estimated from flat residues was used as an input for Monte-Carlo simulations in MatLab to determine the error in the fitted parameters.

In a TROSY-based experiment the exchange free component relaxes with the rate R₂₀β where R₂₀β=R₂₀–η_x+R₁^H/2. η_xy is the cross correlation between the ¹⁵N-¹H dipolar interactions and the ¹⁵N chemical shift anisotropy, which can be measured by determining the ratio of peak intensity for the I_β/I_α using a pulse sequence developed by Wang et al.⁶⁴ R₁^H can be estimated through R₁^H~R₁^HzNz-R₁^N using both a TROSY pulse sequence to calculate R₁^HzNz and a standard experiment to measure the longitudinal relaxation on each amide nitrogen (R₁^N).⁶⁵ The combination of these three experiments allows one to calculate the independently determined R₂₀β.³¹

The exchange-free transverse relaxation time of the TROSY component was determined using pulse sequences designed to measure R₂^α and R₂^β with a 2τ delay of 43.2 ms to ensure a ratio I^β/I^α~0.3 for most residues, and R₁^2HzNz with a τ delay of 21.6 ms.^{31;64 15}N R₁ values were measured using standard experiments⁶⁵ with the following T₁ delays: 0.01, 0.1, 0.5, 0.7, 1.1, 1.5 s. R₁^H values were measured using a TROSY-based R₁^HzNz pulse sequence (L.E. Kay, personal communication) employing a reburp pulse to selectively invert the amide region with following delay times: 0.025, 0.04, 0.06, 0.1, 0.15, 0.2, 0.3, 0.35 s. Data was processed using NMRPipe. Intensities were fit to a mono-exponential decay curve, and uncertainties were measured from duplicate points and an estimate of signal-to-noise (2%).

³J_NCγ couplings for aromatic and residues were obtained from quantitative J-correlation spectroscopy³⁶ in duplicate with a dephasing time of 50 ms.

ATPase activity assays

Full-length dimeric WT and Y101L NtrC with 3 alanine mutations in the c-terminal domain to prevent higher order oligomerization upon activation⁶⁶ were grown as described previously.⁶⁶ Cells were resuspended in buffer (50mM TrisOAc, pH 8.2, 50mM KCl, 10% v/v glycerol, 0.02% NaN₃, and 0.4mM TCEP) and lysed using 1mg/mL lysozyme (Sigma) at 25°C for one hour. Cleared lysates were precipitated by a 0-35% (NH₄)₂SO₄ saturation and were further purified by ion exchange chromatography using a Q Sepharose Fast Flow column (50 mM TrisHCl, pH 7.5, 20 mM NaCl, 0.4 mM TCEP) with a 1M NaCl gradient over 8 column volumes for elution. Rates of ATP hydrolysis were determined using a colorimetric malachite green assay⁶⁷ in order to detect released inorganic phosphate (linear range 0.2-40 μM). Reactions were carried out in triplicate in a 96 well plate using malachite green reagents (Bioassay Systems) and KH₂PO₄ as the standard. Briefly, 1 μM final dimeric NtrC was mixed in with assay buffer (50 mM HEPES·KOH, pH 7.3, 100 mM KOAc, 8 mM Mg(OAc)₂, and 5 mM β-ME) in the presence and absence of 200μM BeCl₂ and 5 mM NaF and incubated at room temperature for 5 minutes. Following the addition of 0.1mM equimolar Mg²⁺ATP, pH 7.0, 80 μL aliquots were taken at 30 second intervals and quenched in 20 μL of the reaction solution (0.337 g Malachite green in 750 mL of HPLC grade H2O combined with 10.5 g ammonium molybdate (Sigma) in 250 mL of 4N HCl). The formation of the reactant malachite green/molybdate/Pi complex was measured by absorbance at 620 nm after incubation at room temperature for 30 minutes to allow for color development.

Molecular simulations

The active structure of the receiver domain of NtrC, refined in a previous study in our lab,⁶⁸ was used as starting structure for the simulation. The system was parameterized with the CHARMM 22-protein all-atom force field with the CMAP backbone energy correction included^{69; 70} using the psfgen plugin in VMD.⁷¹ The system was solvated in a rectangular box with TIP3 water molecules and neutralized with 7 sodium ions. The simulation box therefore contained approximately 24,600 atoms. Periodic boundary conditions were applied to the simulation box. After energy minimization, the simulation box was gradually heated to 310 K with a time step of 1 fs while gradually releasing positional restraints in a MD simulation of 2 ns with the software NAMD 2.7b2.⁷² The system was equilibrated for an additional 70 ns in the NPT ensemble (T = 310 K, P = 1.01325 bar) with the software NAMD, using the Langevin dynamics method⁷³ for temperature control, and the combined Langevin piston Nose-Hoover method^{74; 75} for controlling pressure.

Further testing and the production run were then performed on the supercomputer Anton, designed and produced by DESRES research.⁷⁶ The final configuration of the NAMD trajectory was converted to run on Anton using a set of scripts provided by the staff at PSC and tests were performed to evaluate performance and accuracy and to tune the parameters for the run. The final parameters used for the production runs on Anton were a timestep of 2.5 fs, a 32×32×32 FFT mesh and a 1:1:2 RESPA scheme. With those parameters, the energy drift rate in test NVE simulations was approximately 0.017 Kcal/mol/Dof/us, which for our system (~ 50000 dof) matches DESRES published results for similar systems.⁷⁷ In all subsequent runs the systems were coupled to a Nose-Hoover thermostat.

The performance of the runs was about 12 μs/day. After a first exploratory run at 310K (data not shown) for ~12 μs the system was reintroduced to NAMD, re-equilibrated to 320K, reconverted to Anton and simulated at 320K for approximately 71.5 μs.

The pseudoangle used for describing the rotation of the helix along its axis in Fig 3C is defined in⁶⁸ as the dihedral angle identified by the position of the four center of mass of the following atom groups:

• Group1: residues Val8, Val9, Thr32, Phe33, Leu52, Ser53, Ile80

• Group2: residues Leu87, Asp88, Ala89, and Ala90

• Group3: residues Val91, Ser92, Ala93, and Tyr94

• Group4: residues Asp88, Val91 and Ser92

Δδ¹⁵N were computed for Fig. S2 and Fig. S3 with Sparta+³⁷ for the three possible exchange processes: t ⬄ g−; g+ ⬄ g−; and average t,g+ ⬄ g−. For each of the single states (t, g+ and g−) the chemical shifts have been averaged over 5000 frames extracted from the simulation.

Highlights.

• NMR dynamics show a second process that is faster then the activation transition

• This process is related to χ₁ rotamer exchange of Y101

• MD simulations show that Y101 motion is uncorrelated with the activation transition

• Experimental data against the dominant ‘Y-T coupling’ mechanism of activation in NtrC