Sequence determinants of allosteric back-to-front control of the Arf nucleotide switch

Abstract

Unlike AI-based protein structure prediction, the sequence determinants of protein dynamics, and thus function, remain elusive. The nucleotide switch in Arf GTPases involves a massive structural change, which we showed recently in Arf1 is facilitated by a dynamic molten globule ensemble.

Here we investigate the unresolved sequence-dynamics paradigm by comparing Arf1 and Arf6 using a combination of high-pressure NMR and other biophysical methods. We show that, as for Arf1, the Arf6 nucleotide switch implicates a functional molten globule ensemble, suggesting that this mechanism is a general feature of Arf and Arf-like GTPases. Comparison of the local stability distributions identifies key sequence determinants for the differences in switching between Arf1 and Arf6. Remarkably, these determinants are distant from the nucleotide-binding site, revealing back-to-front allosteric coupling controlling the switch. Evolutionary covariance analysis suggests that this back-to-front allosteric coupling is a fundamental characteristic of the Arf and Arf-like family, and may extend to the entire small GTPase kingdom. Our study thus establishes notable sequence-dynamics relationships with implications for signaling and diseases.

Graphical Abstract

Introduction

While AI-based algorithms have been highly successful in predicting protein structure from primary sequence [1–4], the sequence determinants of protein dynamics, and thus their functional properties, remain to be determined [5]. Lack of understanding of the relationships between sequence and dynamics represents an important knowledge gap, whether for understanding function, for targeting proteins in human disease, or in protein design for a broad range of biotechnological applications. Model systems in which functional sequence/dynamics relationships can be interrogated are thus of primary importance in filling this knowledge gap. The ADP-Ribosylation Factors (Arfs), a class of small GTPases within the Ras superfamily which is responsible for regulating vesicle trafficking and membrane organization [6–8], represent such a system. Like all small GTPases, Arf GTPases function as molecular switches by toggling between a GDP-bound ‘off’ conformation and a GTP-bound ‘on’ conformation (Figure 1) [9]. However, Arf GTPases and related Arf-like (Arl) family proteins undergo a massive conformational rearrangement during their GDP/GTP switch (Figure 1) that is not observed for other small GTPase families [10]. We showed recently that in Arf1, the most abundant Arf family member, this large structural change involves considerable conformational dynamics which is facilitated by a molten globule ensemble [11]. Thanks to their high homology and substantial conformational dynamics, comparison of homologous Arf GTPases presents advantages for the elucidation of the sequence determinants of functional dynamics.

Figure 1. Arf6 structures.

A) Arf6 GDP pdb code 1e0s [24]; B) Arf6 GTP pdb code 2j5x [29]; C) Cavities within Arf1-GDP. In all panels colors are as follows: the N-terminal region (green) switch 1 (red), interswitch (yellow) switch 2 (light magenta) and C-terminal region (wheat). The nucleotide is shown in CPK sticks, Mg²⁺ is shown as a magenta sphere. Cavities were calculated in Pymol as described in Methods using a detection radius of 3 solvent radii and a detection cutoff of 2 solvent radii.

Arf1 and Arf6 are the best studied Arf GTPases, and are highly homologous in both sequence and structure (>95% similarity)[12]. Arf1 is primarily implicated in vesicular trafficking between cellular organelles while Arf6 functions at the crossroads of trafficking, cytoskeletal and signaling pathways [6,13–18]. Reflecting their central role in the biology of the cell, mutations in Arf regulatory and downstream partners have been linked to a variety of diseases such as cancer, neurodegenerative diseases, or infections [8,19–23].

The GDP/GTP switch in Arf1 and Arf6 have similar structural characteristics. When GDP-bound, Arfs adopt an auto-inhibitory conformation in which the N-terminal helix, an extension with respect to the canonical GTPase module, is anchored into the protein (Figure 1A). In this conformation, the nucleotide-binding switch 1 and 2 elements are remote from the nucleotide, and the nucleotide-binding site cannot bind GTP due to a retracted conformation of the interswitch [24–26]. In this ‘off’ conformation, the protein is cytosolic and does not act to recruit its downstream partners. Conversion to the GTP form first requires that the autoinhibitory N-terminal helix be displaced, which occurs through its interaction with a membrane [27]. In addition, the switch 1 β-strand (Figure 1A, B, red) dissociates from the central β-sheet and moves several angstroms to eventually interact with the GTP ligand, the interswitch β-hairpin (yellow) undergoes a 2-residue register shift upward to occupy space vacated by the N-terminal helix, and the switch 2 region (violet) becomes more ordered and moves towards the nucleotide [28–32]. Due to this autoinhibitory state, spontaneous switching in Arfs is very slow. Efficient GDP/GTP alternation in Arfs requires guanine nucleotide exchange factors (GEFs) for the GDP to GTP switch and GTPase activating proteins (GAPs) for GTP hydrolysis [33,34].

This massive conformational change reshapes most of the N-terminal half of the protein, which includes the interswitch, buried in the protein core. Crystal structures of Arf1 have only captured the interswitch in the “retracted” or “shifted” conformation (Figure 1A, B), (e.g., [25,26,28,30,31]). However, these structures are not sufficient to explain how Arf1 remodels its inner core structure without completely unfolding. We previously used high pressure perturbation (HP) coupled with NMR, SAXS, fluorescence and molecular dynamics simulations to characterize the energetics and structural properties of the excited conformations populated transiently as the switch of Arf1 occurs [11]. These studies revealed that Arf1-GDP populates a classical dynamic molten globule (MG) ensemble, a state previously described for protein folding [35]. This MG ensemble facilitates its nucleotide switch transition, and thus, these conformational dynamics are implicated in function. In addition, we found that the α5 helix, which is located in the C-terminal half of the protein, distant from the switch and which does not undergo conformational changes, was the most stable element in the Arf1-GDP structure. We proposed that it imposes energetic constraints against switching in Arf1 [11].

Our findings suggested that a functional dynamic MG ensemble may support the GDP/GTP switch in other Arf and Arf-like proteins. However, whether individual MG ensembles have determinants that contribute to functional specificities is not known. Here, we focused on the comparison of Arf6 and Arf1. Their structures are nearly identical as are their interfaces implicated in interactions with regulators and effectors [34,36]. Yet, despite these similarities, they exhibit distinct cellular functions. Arf6 also presents biochemical differences with respect to Arf1. Under physiological conditions, spontaneous nucleotide exchange is fully auto-inhibited for Arf1, while Arf6 undergoes slow, yet detectable, spontaneous exchange. A weaker affinity of Arf6-GDP for Mg²⁺ encoded by a single residue change was proposed to contribute to this difference [24] and possibly to an increased flexibility of the switch regions [37]. Another striking difference is the differential sensitivity of Arf1 and Arf6 to small molecules that inhibit their activation by GEFs [38–40]. Notably, Arf6 is insensitive to inhibition by Brefeldin A, a widely used uncompetitive Arf1 inhibitor, despite the fact that all residues of Arf1 in contact with the drug are identical in Arf6 and have the same conformation in their GDP-bound forms [31,38,41].

Together, these functional and mechanistic differences suggest that Arf1 and Arf6 may sample distinct conformations during the nucleotide switch that are perceived by nucleotides, inhibitors, regulators or downstream partners. In the present work, we sought to establish the sequence determinants of the distinct functional properties of Arf1 and Arf6 within the functional MG framework. Using HP coupled with NMR, SAXS, fluorescence and molecular dynamics simulations, we demonstrate that the switching mechanism of Arf6 likewise implicates a functional MG ensemble. However, the Arf6 MG is energetically distinct from that of Arf1. Notably, we show that helix α5 in the C-terminal backside of Arf6 is less stable than that of Arf1, and that mutating the α5 helix in Arf6 to that of Arf1 both stabilized this region and slowed nucleotide exchange. Thus, a novel “back-to-front” allosteric mechanism tunes the nucleotide switch differentially between Arf1 and Arf6. Evolutionary covariance analysis suggests that this “back-to-front” allosteric tuning of nucleotide switching may be a general feature of small GTPases of the broader Ras family. Furthermore, it is conceivable that an equivalent allosteric mechanism may operate in Rag GTPases, which are evolutionarily related to Arfs [42], to support their activation by GPCRs and the mTOR complex, respectively. This knowledge opens new possibilities for allosteric modulation of these GTPases by small molecules to combat disease.

Results

A functional molten globule ensemble is a major feature of the Arf6-GDP conformational landscape

Arf6 populates invisible, dynamic and heterogeneous excited states under pressure

The crystal structure of Arf6-GDP [24] contains internal cavities (Figure 1C), indicating that Arf6 would be amenable to pressure perturbation as a means of exploring its conformational landscape [11,43,44]. Coupling HP with NMR provides residue-specific information on biomolecular stability and thus represents a powerful tool to map protein energetic and conformational landscapes. To characterize Arf6 excited states, we first carried out NMR ¹H-¹⁵N HSQC measurements as a function of pressure. Since Arf6 was purified as a mixture of both the GDP and GTP-bound forms [45] (Figure S1), GDP loading was performed [37] prior to HP NMR experiments to obtain a single set of peaks corresponding to pure Arf6-GDP. In addition, because the backbone amide assignments of Arf6-GDP were unavailable, assignment of Arf6 (140 of the 175 residues, ~80%) was performed using ¹⁵N and ¹³C labeled Arf6-GDP and standard triple resonance experiments (Table S1). The majority of unassigned residues were either proline residues or residues located in the switch regions. Low HSQC peak intensity of many switch residue amide peaks suggests that these regions are inherently labile.

Pressurization of ¹⁵N-labeled Arf6-GDP led to decreased intensity and chemical shift perturbations for the native state amide cross peaks of the ¹H-¹⁵N HSQC spectrum of Arf6 (Figure 2A, B). Chemical shift perturbations as functions of thermodynamic parameters such as heat and pressure have been established for decades [46]. Linear pressure-dependent chemical shift perturbations result from compression of the amide bond, whereas non-linear chemical shift perturbations are indicative of the population of excited states within the native state ensemble [46–49]. To map these excited states onto the Arf6 structure, pressure-dependent changes in chemical shifts were fit to a second-order polynomial [47,50,51], with the first coefficient corresponding to the linear compression coefficient, and the second to the non-linear, conformational change coefficient [50,52]. The magnitude of the non-linear coefficients for both ¹H and ¹⁵N amide chemical shifts of Arf6 were similar to those observed for RalGDS [51], a particularly labile protein. For Arf6, the non-linear coefficients were largest for residues in the N-terminal half of the protein (through switch 2) and in the C-terminal helix (Figure S2A, B), even after correction for the non-linear shifts observed for the corresponding amino acid in a small peptide [50]. Thus, Arf6 populates excited states within the native Arf6-GDP manifold implicating these regions of the protein.

Figure 2. Pressure dependence of amide HSQC spectrum of Arf6-GDP.

A) ¹H-¹⁵N HSQC Plots at atmospheric pressure (red) and 2400 bar (indigo). B) Pressure dependence of Arf6-GDP chemical shifts from atmospheric pressure (red), 400 bar (orange), 800 bar (gold), 1200 bar (green), 1600 bar (blue), 2000 bar (purple), and 2400 bar (indigo).

In addition to undergoing pressure-dependent shifts, the intensities of the native state amide peaks of Arf6 decreased without the appearance of new resonances corresponding to the unfolded state (Figure 2), indicating that Arf6 does not unfold in the range of pressures from 1 to 2,400 bar. This loss of native state peak intensity was ~85% reversible, indicating that protein aggregation was not significantly responsible. Measured peak volumes decreased concomitantly with peak intensities, indicating that peak broadening did not contribute significantly to the decrease in native state amide peak intensity (Figure S2C, D). Rather, this loss of intensity without the appearance of new peaks indicates the population of invisible excited states in intermediate timescale conformational exchange with the native state and/or among themselves.

The pressure-dependent intensity profiles for 132 of the 140 assigned residues could be quantified. The pressure midpoints (p_1/2 values) for the intensity loss transitions (Figure 3A-G) were broadly distributed between 500 and 2500 bar, indicating the population of a heterogeneous ensemble of states as pressure is increased. Fits of these profiles to local two-state transitions (native to non-native) (Figure 3A-G, lines) yielded apparent local free energies $\left(\Delta G_{app}\right)$ and volume changes $\left(\Delta V_{app}\right)$ for each quantified backbone amide (Figure 4A, B, Table S2). Fits of the individual curves can be found in Figure S3. Local stability was slightly larger for residues in the C-terminal half of the protein than in the N-terminal half (Figure 4A, B). The average $\Delta G_{app}$ for all residues was 2.39 ± 0.81 kcal/mol, while the average $\Delta V_{app}$, was 57 ± 15 mL/mol (Table S2). Thus, at HP Arf6 populates a broad ensemble of invisible excited states, dynamically interconverting on the NMR timescale, and these states are between ~1–4 kcal/mol less stable than the native state.

Figure 3. Pressure dependence of the intensity of the amide HSQC peaks of Arf6-GDP.

A-G) Normalized intensity profiles from A) The N-terminal region (1–33), B) Switch 1 (34–48), C) The Interswitch (49–64), D) Switch 2 (65–86), E) Residues 87–130, F) Residues 131–175. G) A combined plot showing representative fits from each portion of Arf6-GDP. H) Fractional contact map of Arf6-GDP at 1800 bar calculated using contacts from the SMOG server and the intensity values for the amides of each native value at 1800 bar as indicated in the Methods section. Contacts are colored according to the fraction of native intensity. Raw data were fit to a two-state transition curve and normalization was carried out on the data and the fit.

Figure 4. Pressure effects on residue-specific stability, nucleotide binding and exchange of Arf6-GDP.

A) The apparent free energy distribution of Arf6-GDP by residue as obtained from non-linear regression of the pressure-dependent NMR HSQC peak intensity data. Error bars represent the estimated uncertainty on the values of each ΔG_app. B) Histogram of ΔG_app values. Residues 1–86 are in red, and residues 87–175 are in blue. C) Pressure effects on the ³¹P spectrum of Arf6-GDP. Arf6-GDP at atmospheric pressure shows two broader peaks for the α and β phosphate groups, corresponding to bound GDP (red). As pressure increased (yellow, green, aqua and purple), the positions of the peaks shift towards the values for free nucleotide complexed with magnesium as previously reported (53). At 2400 bar (blue), peaks at the position of the free nucleotide (~−6.5 and ~−10.5 ppm) are observed while those of the bound nucleotide decrease in intensity. D) Tryptophan fluorescence-detected nucleotide exchange kinetics for Arf6-GDP as a function of pressure, as indicated.

Pressure promotes switching

Next, we asked whether the excited states that appear under pressure display nucleotide exchange perturbations. First, we analyzed the interactions of GDP with Arf6, using ³¹P NMR as a function of pressure (Figure 4C). At atmospheric pressure, two ³¹P peaks distinct from those of free GDP-Mg²⁺ are visible, corresponding to Arf6-bound GDP α- and β-phosphates. With increasing pressure, the position of the β-phosphate peak shifted significantly down-field, while the change in the α-phosphate peak was minimal. The dramatic shift of the β-phosphate of the bound GDP compared to the small movement of its α-phosphate suggests that pressure primarily perturbs the environment of the β-phosphate. At 2400 bar, in addition to the down-field shifts, the intensity of the bound peaks decreased, while concomitantly, a second set of narrower peaks became apparent. These new peaks align with those of free GDP-Mg²⁺ at this pressure, −6.5 and −10.5 ppm (also see [53]), indicating that above 2000 bar, pressure leads to dissociation of a significant fraction of the GDP from Arf6. Note that in prior work [11] we had assumed that had the nucleotide dissociated, it would have corresponded to free GDP, rather than free GDP-Mg²⁺, which is what is actually present in out buffer conditions (50 mM bis-Tris, 5 mM DTT, 150 mM NaCl, and 1 mM MgCl₂, pH 6.5). No nucleotide dissociation was observed in those prior studies, hence our conclusions stand. However, the free nucleotide spectrum shown for comparison sake in that study should have been that of GDP-Mg²⁺, rather than fere GDP, which exhibits a pKa that is shifted to higher pH values [53].

We then examined nucleotide exchange kinetics as a function of pressure using tryptophan fluorescence. GDP/GTP nucleotide exchange manifests as a fluorescence increase, and GTP hydrolysis as a fluorescence decrease [39]. Spontaneous exchange of Arf6-bound nucleotide occurred at atmospheric pressure in relatively low Mg²⁺, 1 μM, (Figure 4D, red), as previously observed [24], with the onset of GTP hydrolysis also apparent at later time. The complex and multistep character of the exchange transition precludes here a full mechanistic analysis, but compared to the kinetics at atmospheric pressure, the rate of nucleotide exchange increased significantly at 1500 bar (Figure 4D, green), indicating that the excited states facilitate nucleotide exchange.

Arf6 remains globular and retains its secondary structure under pressure

While NMR can provide insights into pressure-induced changes in the environment of specific residues, small angle x-ray scattering (SAXS) yields information about a protein’s global size and shape. At 1 bar, Arf6 was globular, consistent with its crystal structure, with a radius of gyration, $R_{\text{g}}$, of ~ 18.2 Å and a pair distance distribution (P(r) plot) corresponding to a spherical protein with a small extension (Figure 5A, B). The normalized Kratky plot was also consistent with a globular shape (Figure 5C). As pressure increased, the $R_{\text{g}}$ value first decreased slightly to ~17.6 Å at 1000 bar, then increased to 19.2 Å at 2,500 bar (Figure 5D). A larger fraction of pairwise distances between 40–50 Å was apparent in the P(r) plot at high pressure, while the extension $\left.\left(D_{\max }\right.~67\text{Å}\right)$ was maintained throughout the pressure series (Figure 5B). The normalized Kratky plot indicated a somewhat less spherical, yet still largely globular protein at high pressure (Figure 5C). HP-FTIR analysis revealed that in addition to remaining globular at high pressure, Arf6 mostly retained its secondary structure (Figure S4).

Figure 5. High Pressure SAXS measurements of Arf6-GDP at pH 6.5.

A) Scattering intensity profiles collected at 10 (red), 500 (green), 1500 (blue), and 2000 (purple) bar. B) Pair distance distribution functions calculated from the intensity profiles over the same pressure range. Color code is noted. C) Normalized Kratky plots calculated from the intensity profiles at 10 (red) and 2000 (purple) bar. D) Radius of gyration as a function of pressure, determined from the P(r) analysis and taken as an average of four separate experiments. Error bars represent standard deviations from the mean of the 4 experiments.

A broad basin of excited states exists on the Arf6 landscape

A conformational landscape for Arf6 was derived using structure-based modeling [43,44] and the pressure-dependent NMR profiles as constraints. To do so, fractional contacts for each pair of native contacts in Arf6-GDP at each pressure were calculated, as previously [11,43,44], as the geometric mean of their fractional intensities in the NMR profiles (see Methods section, Figure 2H, Figure S5). Then for each pressure, 100 Cα coarse-grained simulations were performed in which each of the contacts was present randomly across the 100 contact lists at the fractional contact value for that residue and that pressure [11,44]. The composite conformational landscape obtained from 400, 100 ns simulations based on the fractional contacts at 1, 800, 1200 and 1800 bar exhibited a broad intermediate ensemble between Q ≈ 60–80% (Figure S6A). The switch region in this ensemble becomes progressively less structured with decreasing overall Q-value (fraction native contacts). This ensemble of low-lying excited states is a major feature of the Arf6 energetic landscape.

The conformational landscape (Figure S6) derived from coarse grain modeling based on the NMR-derived fractional contact maps tends to over-represent highly disrupted conformations, as the starting contact lists for the simulations either have a contact present or do not based on the fractional contact value for that residue and pressure. Thus, Ensemble Optimization Modeling (EOM) [54,55] of the HP SAXS data (Figure 5A) was used to constrain the configurational landscape of Arf6-GDP to significantly populated conformers. At atmospheric pressure, the configurations recovered from the EOM modeling using the ensemble at 10 bar were mostly folded globular states (87%), along with a small fraction (9%) of a partially disrupted conformer and a very small fraction (4%) of an extended conformation (Figure S7, Table S3). At high pressure, the fraction of the extended conformation recovered from the EOM modeling increased significantly (from 4% to 11%) (Figure S8, Table S3). In both cases, the agreement between the EOM-derived sub-ensemble and experimental data was quite high, as represented by the low χ² values. No such extended conformations were apparent in prior EOM analysis of HP SAXS data on Arf1-GDP [11].

Thus, the excited state ensemble of Arf6 favored under pressure is dynamic and heterogeneous (NMR), energetically excited (lower stability), mostly globular (SAXS), hydrated (significantly lower molar volume), with significant secondary structure (FTIR). These properties correspond to those of a classical molten globule [35,56–58]. Its population under pressure facilitates nucleotide exchange, and thus this MG ensemble in Arf6 plays a functional role in nucleotide switching.

Back-to-front control of switching is modified in Arf6 relative to Arf1

Arf6 local stabilities are distinct from those of Arf1

We showed previously that Arf1 forms a classical MG ensemble under pressure which favors the GDP to GTP transition, supporting a functional role in the nucleotide switch [11]. Our above data now extends this MG-based mechanism to Arf6 function. Given the known biochemical differences between Arf1 and Arf6, we thus asked whether these differences can be linked to differences in the characteristics of their respective MG ensembles. As noted above, Arf6 exhibits 83% sequence similarity and 69% sequence identity to Arf1 (Figure 6). The majority (2/3) of sequence differences are located in the C-terminal half, which does not contain switch elements and is located at the “back” of the GTPases, opposite to the nucleotide-binding site and the switch.

Figure 6. Sequences differences between Arf6 and Arf1 are most significant in the C-terminal half, on the “backside” of the protein.

A-C) Side, back and front views of the overlay of the Arf1-GDP (light green) and Arf6-GDP (light blue) structures bound to GDP. Substitutions in the switch regions and in helices α4 and α5 are shown as red sticks. D) Sequence alignment of Arf1 and Arf6. The regions encompassing switch 1, interswitch, switch 2 and helices α4 and α5 are indicated by colored and labeled bars.

We first compared the global and individual stabilities of Arf1 and Arf6 (Figure 7 A, B, Table S4). Surprisingly, Arf1 and Arf6 have very similar overall apparent stability (Figure 7B, [11]), indicating that the faster nucleotide exchange kinetics in Arf6 do not result from a global stability loss, but must have another origin. In fact, the N-terminal switch region in Arf6 was found to be slightly more stable than in Arf1, while on average the C-terminal half of Arf6 was similarly stable. Interestingly, despite similar average stability, some residues in the C-terminal half of Arf6 were more stable than in Arf1, while others were significantly less stable. Notably in helix α4 (residues 96–112) and the kinked helix, α5 (residues 131–145) (Figure 7A, B, Table S4) the local stabilities of several residues was over 1.5 kcal/mol lower for Arf6. This value of 1.5 kcal/mol is larger than the largest values of 2σ for the values of ΔΔG(Arf6-Arf1). Interestingly, helices α4 and α5 contain the largest fraction of substitutions between Arf1 and Arf6 (Figure 6).

Figure 7. Stability Differences between Arf6 and Arf1 and the role of helix α5.

A) The difference in the residue-specific $\Delta G_{app}$ values, $\Delta \Delta G_{app}$, between Arf6 and the equivalent residue in Arf1 (numbering is Arf1). Red rectangle highlights residues in Arf6 that are markedly less stable than in Arf1. B) Average $\Delta G_{app}$ values for various regions of Arf1 (light green - taken from (11)) and Arf6 (light blue). C) Average stability of all quantified residues, left, and of helix α5, right, of Arf6α5 (blue) relative to WT Arf6 (red). D) Spontaneous GDP-GTP Exchange at 1 μM MgCl₂ and at atmospheric pressure from WT Arf6, Arf6α5, Arf6-KPH and Arf6 P132A measured by tryptophan fluorescence. Equimolar GTP was added to the solution. Colors are as noted in the figure.

Mapping these substitutions in the 3D structure of Arf6-GDP suggests that they are important for the loss of stability in this region, with a small number of stabilizing interactions compensated by a larger number of destabilizing interactions (Figure S9). More stable elements in Arf6 include a new hydrogen bond in Arf6 between D97 and Q101 in the middle of α4 (Figure S9B) (~1 kcal/mol) and an increase in stability in ¹⁴³LGL¹⁴⁵ in the middle of α5 (~1 kcal/mol). Conversely, H-bond distances increase by 0.5 Å between E115 (111 – Arf6 numbering) and the backbone nitrogen of residue 84 (80 in Arf6) (Figure S9A). Substitution of A136 in Arf1 by P132 in Arf6 disrupts the H-bond pattern in helix α5 (Figure S9C), such that the majority of residues in helix α5 (P132 was not measured), exhibited a decrease in stability in Arf6. Notably, the reorientation of the side chain of T142 (Arf6) with respect to H146 (Arf1) disrupts packing with the side chain of the W149 (W153 in Arf1) in an adjacent strand, resulting in a strong destabilization (1.2 kcal/mol) at this position (Figure S9D). In addition, about half the residues in helix α4 were found to be less stable in Arf6 (Table S6). Overall, the lower stability of residues in helix α5 in Arf6 leads to the disappearance of the relatively stable intermediate at Q ~ 35% observed on the simulated Arf1 configurational landscape (Figure S6B) [11], which is mostly unfolded but retains secondary structure in helix α5.

Arf1 mutations in helix α5 of Arf6 lead to its stabilization and slower switching

We proposed previously that helix α5, because it is the most stable element in Arf1, acts as an allosteric belt to constrain switching in Arf1 [11]. We therefore sought to establish whether the differences in sequence and stability between Arf1and Arf6 in helix α5 play a role in their distinct switch properties. We first tested the stability of the complete α5 mutant of Arf6, which replaced the entire sequence from positions 131 to 143 in Arf6 with the Arf1 sequence (a total of 7 amino acid substitutions). To perform local energetic mapping by HP NMR we first verified and modified assignments using standard triple resonance approaches (Table S5). The HP NMR results revealed that, while the global stability of the protein was not significantly altered (Table S6, Figure 7C), the residues constituting the α5 helix were markedly more locally stable in the mutant, increasing by 0.87 kcal/mol on average, a 37% increase in stability of this region. Thus, the Arf1 sequence is sufficient to render α5 more stable in the Arf6 environment.

Next, we established the nucleotide exchange kinetics of the 7-residue Arf6 α5 mutant. Remarkably, nucleotide exchange kinetics at atmospheric pressure and 1 μM Mg²⁺ were noticeably slower than for WT Arf6 (Figure 7D), with the time to full exchange increasing from ~3600 seconds in the wild-type Arf6 to greater than 8000 seconds. To refine the role of individual sequence differences, two other Arf6 mutants were tested; one in which the residues ¹³¹KPH¹³³ in Arf6 were substituted with NAA from Arf1, as well as the single P132A mutant. Both variants also exhibited slower nucleotide exchange kinetics, with the time of full exchange greater than 8000s. Interestingly, the KPH mutant exhibited a slower exchange rate than either other mutant, suggesting that while mutation of a single proline is sufficient to significantly impact exchange, this effect is enhanced with the substitution of the flanking residues. In contrast, the 7-mutation variant reverted to the overall exchange rate observed with the P132A mutant, implicating compensating effects by the further substitutions.

Taken together, these results confirm that the stability of helix α5 modulates Arf protein nucleotide exchange rates. Thus, our data highlight the existence of a novel allosteric control of the switch regions by elements located at the back side of the protein, which exhibit significant sequence variance between the two homologs. We refer to this heretofore unknown allosteric modulation as “back-to-front” control of switching.

Structurally distant evolutionary couplings in the C-terminal “back side” of Arfs suggest back-to-front control of switching extends across the small GTPase superfamily

Since helix α5 is structurally conserved, although not always kinked, in small GTPases (reviewed in [9]) and is not a switch element, this raised the question of whether the back-to-front allosteric mechanism is operative in other members of the broader Ras family of small GTPases. For deeper insight into the energetic networks in small GTPases, we carried out evolutionary covariance analysis using the Direct Coupling method available at the EV Couplings web server [59– 62]. In a multiple sequence alignment (MSA), evolutionary coupling is observed generally for residues in direct contact in the three-dimensional structure of homologues, but can also arise due to subunit interactions, distinct conformational states, and allosteric networks. Excellent covariance scores (Q = 9–10) were obtained for three different depths of the MSA (Figure S10A) using the Arf1 sequence as bait. Results were similar when the Arf6 sequence was used (Table S7). Interestingly, few of the highly probable couplings were found in the switch region. The fact that the switch regions largely do not contain evolutionarily coupled residues may reflect their high level of conservation across the super-family, and otherwise that the evolution of these residues is dependent on interactions with binding partners and thus is not accounted for in analysis of the monomer.

We were particularly interested in residue pairs that exhibit high coupling probability but are distant from each other in the 3D Arf structure (Structurally Distant Evolutionary Couplings, SDECs), originally referred to as false positive evolutionary couplings in the context of protein structure prediction (See example in Figure S11A)[59,60]. Because Arf GTPases have two distinct end-point conformations, we excluded from our analysis residue pairs that formed SDECs in one nucleotide-bound state but a direct contact in the other state (Figure S11B). Bonafide SDECs, residues, which do not form contacts in either of the nucleotide-bound states, are indicative of allosteric communication or requirements for coupled local stability. Remarkably, these very high probability (>94%) SDEC pairs were found to constitute an extensive network of residues in the C-terminal half of these proteins (Figure 8, Figure S10, C,D, Table S7), notably including helix α5. The existence of these coupled networks thus suggests that “back-to-front” control could play a significant role in tuning switching across the large family of Ras GTPases.

Figure 8. Structurally distant evolutionary couplings in small GTPases.

A) True Structurally Distant Evolutionary Coupled (SDEC)residues in the C-terminal region of Arf1, the bait sequence for the evolutionary covariance analysis. Distances between co-varying residue pairs are all > 9 Angstroms in the GDP-bound (A), as well as in the GTP –bound (B) structures. The view is of the back of the protein, with the switch region behind.

Discussion

In this work, we have demonstrated that Arf6 populates a canonical molten globule ensemble that is stabilized by high pressure and facilitates switching. This MG ensemble is of higher energy than the native Arf6-GDP state (1–5 kcal/mol), heterogeneous (broad local stability distribution), globular (minimal changes in R_g) and hydrated (significant molar volume changes). Its conformations, which retain their secondary structure (as per HP FTIR), are in dynamic exchange on the NMR timescale. These characteristics explain how Arf6 is able to reorganize its core structure upon switching without complete unfolding. Because activation of Arf GTPases by their GEFs is a multistep mechanism involving a similar reorganization of the core β-strands [28,31,41], it is likely that conformations within the MG ensemble are recognized and exploited by Arf6GEFs during the physiological switch. We previously characterized a similar MG ensemble for the Arf1 homolog [11], supporting the notion that the MG ensemble plays a general role in Arf family function. The Arf and Arf-like small GTPases form a separate evolutionary branch within the Ras family of small GTPases [10,42]. This branch has evolutionary relationships with the Rag GTPases, which are major components of the mTOR megacomplex and exhibit a roadblock domain C-terminal to the GTPase domain [42]. Arf-like and Rag GTPases, as for Arf proteins, display a massive switch that reorganizes the core of the GTPase, and hence must resolve the same problem as Arf GTPases [10,63]. Our results, demonstrating that the functional MG ensemble exists in two different Arf GTPases, reinforce the generality of MG-based mechanism for GTPases which must undergo a large conformational change during the nucleotide switch.

Our study also reveals that, while globally similar, the MG-based functional mechanisms for Arf1 and Arf6 have specific sequence determinants that encode specific functional properties in Arf isoforms. First, the differences in the energy landscape of Arf6 (this study) compared to that of Arf1 [11] alter the population and structures of conformations within the MG ensemble. These energetic and conformational differences may be important for their discrimination by regulators and effectors. Notably, as GEFs remain bound to Arf GTPase throughout the GDP-GTP switch, we propose that they discriminate between, and possibly stabilize, specific conformations within the MG ensemble.

Finally, we found that the C-terminal half of Arf6, in particular helix α5, is much less stable than in Arf1. This helix has a large fraction of the substitutions between the two homologues, although the implications of these differences remained to be determined. Interestingly, although helix α5 is located outside the switch elements and remote from the nucleotide-binding site, we demonstrated that it facilitates nucleotide exchange in Arf6 (this study) while it opposes the switch in Arf1 [11]. These observations reveal a novel allosteric modulation of the switch in Arf GTPases, distinct from the original canonical allosteric control of the nucleotide-binding site by the autoinhibitory N-terminal helix in the Arf family [10]. Importantly, our study establishes that a small number of sequence determinants are critical for dynamics and functional specificities, and reveals that such determinants may be located outside the reactive regions of the protein.

The allosteric mechanism brought to light in the present work emanates from a region that does not contact GEF and GAP regulators [34], raising the question of whether, in addition to modulating switch kinetics, it can itself be modulated by other intermolecular interactions. Interestingly, the back side of Arf was recently shown to be involved in intersubunit interactions in several large assemblies whose structures were determined by cryo-electron microscopy. In COP1-coated vesicles in complex with ArfGAP, Arf1 molecules located on the 3-fold axis of the asymmetric unit interact with several COP subunits and with ArfGAP, of which a large contact with γ-COP involves helices α5 and α6 [64]. Alternative contacts from the back side are also seen in a Arf1/COP1/NEF assembly, in which Arf proteins form a symmetrical dimeric interface via the C-terminal regions [65]. Extensive contacts from the back side of Arf are also seen in an Arf1/AP1/NEF particle, including the AP1 subunits μ1, the HLA antigen, NEF and the other Arf protein forming a symmetrical dimer, while the switch regions contact only β1-AP [65]. While the details of these interactions are not visible at medium resolution, both assemblies suggest that the C-terminal region of Arf GTPases may have an allosteric function in coordinating coat assembly and/or disassembly. This notion remains to be investigated. In addition, whether this new and the canonical allosteric mechanisms cooperate upon recruitment of Arf GTPases to membranes is an important issue for further studies.

The Ras branch of small GTPases (Ras, Rho/Rac/Cdc42, Rab) differ from the Arf/Arf-like branch in that they do not have an autoinhibitory N-terminal helix that establishes an allosteric control of the nucleotide-binding site and they do not switch through a massive conformational change spanning their inner core. It is therefore intriguing that several highly evolutionary coupled, yet structurally distant residues were highlighted in the C-terminal half of the GTPases, opposite the switch region and the nucleotide-binding site, in a large sequence alignment of small GTPases. This high probability, C-terminal structurally distant residue covariance across a very deep MSA suggests that “back-to-front” allosteric control may extend throughout the larger Ras family. Indeed, structural and/or biochemical evidence consistent with such “back-to-front” allosteric control has been reported. For example, saturation mutagenesis of H-Ras coupled to a bacterial screen in absence of GEFs revealed gain of function (increased activation), non-oncogenic mutations in several residues located in helices α3 and α5 of Ras (α4 and α6 in Arf numbering) that led to increased switching [66]. In particular, mutations at Y157 in α5 increased the rate of spontaneous exchange, and molecular dynamics simulations suggested that there is a loosening of the Y157A mutant structure. This suggests an allosteric effect similar to that of Arf GTPases, in which mutations on the backside of the protein, remote from the switch elements and the nucleotide-binding site reduce the constraints against switching. Interestingly, most of the gain of function mutants were less stable in vivo, consistent with the energetics/switch correlation observed here for Arf1 and Arf6. In addition, diverse amino acid substitutions found in tumors and RASopathies cluster in a pocket centered on helices α3 and α4 in Ras isoforms (α4 and α5 in Arf numbering), a site located at the back of Ras protein that has been proposed to allosterically modulate the nucleotide switch [67,68]. Besides Ras, an oncogenic mutation in helix α3 in Rac1 (N92I) was recently shown to alter the overall stability of the protein, increasing spontaneous switching[69].

In addition to direct control of switching rates, evolutionary covariance in the C-terminal half of small GTPases could implicate functions separate from switching, itself. Notably, protein-protein interactions involving helix α5 (Ras numbering) have been observed for Ras-GTP isoforms in complex with Raf [70,71] and in Ras-GTP bound at the allosteric site of SOS[72]. Moreover, NMR analysis of membrane-attached K-Ras-GTP identified a population of Ras-GTP that interacts with the membrane through helices α4 and α5, including residue Glu 153 which is mutated in a RASopathy, the Noonan syndrome [73]. Dynamic collision of helices α4 and α5 with the membrane was proposed to further extend to Rho-GTP GTPases, using molecular dynamics [67]. It is thus tempting to speculate that members of Ras GTPase branch use back-to-front allosteric communication similar to that described here for Arf1 and Arf6, possibly to inform the nucleotide-binding site on the proximity of the membrane, a hypothesis that will require investigation.

In conclusion, our study provides important insights into the sequence determinants of pivotal functional dynamics in protein signaling. Characterization of the excited states populated during the nucleotide switch is central to understanding Arf family specificity and to developing alternative therapeutic strategies to target these proteins in a variety of diseases (e.g., [8,19–23]). The present discovery of “back-to-front” control in the Arf family of small GTPases and the far-reaching evolutionary relationships uncovered across the broader Ras family warrant detailed investigation of the links between local stability networks and switching across this important group of proteins.

Beyond the small GTPases studied here, our study demonstrates that the HP-based approach is suitable for the elucidation of the sequence determinants of functional dynamics in other protein families. Central to this approach is the ability to map the local stabilities that control conformational dynamics at the level of individual residues using NMR. Coupling HP with other experimental and biophysical methods (SAXS, FTIR, fluorescence, molecular dynamics simulations) allows for structural characterization of the excited states implicated in these functional dynamics. The need for residue-specific resolution from NMR limits the size of the protein systems which can be studied. Future coupling of HP to H/D XMS, while decreasing resolution to the level of small peptides, could extend the local stability mapping to much larger proteins and their complexes.

Materials and Methods

Protein Production and Purification

Arf6 and its mutants were produced as described previously[11]. Unlike with Arf1, Arf6 is not fully bound to GDP when purified from E. coli but rather is bound to both GDP and GTP. Thus, an additional GDP loading step was performed to ensure a single nucleotide-bound state. Purified protein was incubated together with 5 mM GDP and 5 mM EDTA for 1 hour at 37 °C. The loading reaction was then quenched with 20 mM MgCl₂ and immediately subjected to a second round of size exclusion using the same column as previously described [11]. The GDP loaded protein was then concentrated as appropriate for NMR and SAXS measurements. For GTP exchange experiments, this second round of size exclusion was done in an equivalent buffer except with only 1 μM MgCl₂. An equimolar amount of GTP was added to the protein solution and data acquisition began within 2 minutes of nucleotide addition.

High Pressure NMR

Triple Resonance Assignments

NMR data for backbone resonance assignments of Arf6-GDP was obtained using a Bruker Avance Neo 800 MHz spectrometer equipped with a 5 mm TCI HCND cryoprobe with z-axis gradients at 295 K (Rensselaer CBIS Core Facility) using a standard triple-resonance NMR strategy [74]. Additionally, experiments including HNH-TOCSY-HSQC, hC(C)(CO)NH-TOCSY, and Hc(C)(CO)(NH)-TOCSY implemented with non-uniform sampling (NUS) were used to complete assignments. A sampling schedule of 25–50% was used for all NUS experiments and data was visualized with TopSpin 4.2. Data was then reconstructed using Iterative Shrinkage Thresholding (IST) within NMRPipe [75]. Backbone resonance assignments were then determined by manual refinement of the outputs obtained from both the I-PINE web server [76] and the program MONTE [77]. Resonance assignments for Arf6 were deposited in the Bio MagRes Database. For the α5 mutant, assignments of the new and altered peaks were completed with HNCACB experiments. Assignments can be obtained at the BMRB entry number 52984.

High Pressure ¹H-¹⁵N HSQC

¹H-¹⁵N HSQC experiments were performed on either a Bruker Avance III or an Avance Neo 600 MHz spectrometer equipped with a 5 mm TBI HCXD probe with triple axis gradients at 295 K (Rensselaer CBIS NMR Core Facility). High pressures for all experiments were generated using a Xtreme Syringe Pump and 5 mm O.D. zirconia NMR tube (Daedalus Innovations, Aston, PA). Samples contained 0.5 mM protein in buffer (50 mM bis-Tris, 5 mM DTT, 150 mM NaCl, and 1 mM MgCl₂, pH 6.5) ¹H-¹⁵N HSQC spectra were recorded as a function of pressure from 1 bar to 2,400 bar, with a 10-minute equilibration time after a change in pressure given before data acquisition. All NMR spectra were processed using Bruker’s Topspin software and analyzed in either POKY [78], PINT [79], or CCPNMR [80].

High Pressure ³¹P NMR

1D ³¹P spectra were collected on a Bruker Avance III 600 MHz Spectrometer using a 5 mm TBI HCXD probe at 298 K (Rensselaer CBIS NMR Core Facility). NMR samples contained 0.75 mM protein in the previously described buffers. Spectra were acquired using an inverse gated single pulse experiment with a 68.5° (13.37us) tip angle, 80 ppm sweepwidth, acquisition time of 0.210s, 2.6s interscan relaxation delay, 4096 scans, 8 steady state scans, ~14kHz WALTZ 1H decoupling, and transmitter frequency of 243.0249199 MHz.

HP NMR Data analysis

Pressure-dependent chemical shift perturbations in both the ¹H and ¹⁵N dimensions for each residue, $i$, were analyzed using a second order polynomial, as per Akasaka and Li and Kitahara et al [48,49].

$${\partial }_i=a_i+b_{i}p+c_i{p}^2$$ [1]

Recovered second order coefficients, $c_i$, were corrected for the small non-linear effects of pressure on amide chemical shifts of residues in tetrapeptides reported by Koehler et al [50].

For the analysis of the pressure-induced intensity loss transitions, for any given pressure, each residue in a protein ensemble is either in its native state environment or it is not, depending on the change in stability and volume between the native and non-native like conformations. Thus, we fit profiles for the pressure-induced reduction in native state backbone amide ¹H-¹⁵N HSQC peak intensities to a two-state model as follows:

$$I(p)=\left[\frac{I_n+\left(I_n{\text{e}}^{-\left(\frac{\text{Δ}{G}_{app}+p\text{Δ}{V}_{app}}{RT}\right)}\right)}{\left(1+{\text{e}}^{-\left(\frac{\text{Δ}{G}_{app}+p\text{Δ}{V}_{app}}{RT}\right)}\right)}\right]$$ [2]

Here $I_p$ and $I_n$ represent the intensity at pressure, $p$, and the intensity of the native state at atmospheric pressure, respectively. For this analysis, the low-pressure plateau value was fixed as 105% of the average of the first two pressure points, taken at atmospheric pressure and 10 bar. In this equation we constrain the high-pressure plateau, corresponding to 100% population of the excited state, to 0, representing a full depopulation of the native state ensemble and thus no native state peak intensity. The apparent change in the free energies $\left(\Delta G_{app}\right)$ and the molar volume $\left(\Delta V_{app}\right)$ for the transition from native to excited state ensembles was determined on a residue-specific basis by least-squares fitting to the above equation, done with a suite of MATLAB scripts written in-house.

High Pressure SAXS Experiments

High-pressure SAXS experiments were performed at the ID7A beamline at the Cornell High Energy Synchrotron Source (CHESS). Pressure profiles were collected as a function of pressure between 1 and 3,000 bar using a Barocycler HUB440 high-pressure pump (Pressure BioSciences). Samples between 0.3 and 0.5 mM protein in buffer as described above were centrifuged for 15 minutes at 16,000 × g before loading for SAXS measurements. Each acquisition was taken as an average of five 1 second exposures, with a 10-minute equilibration time allowed at each pressure before acquisition. SAXS profiles were analyzed using BioXTAS RAW 2.2.1 [81], and the pair distance distribution function (P(r)) was computed using the GNOM package of the ATSAS 3.2.1 suite integrated into RAW.

High Pressure FTIR Experiments

Lyophilized Arf6-GDP in the previously described 50 mM Bis-Tris at pH 6.5 buffer was dissolved in D₂0 adjusted to pD 6.5 (pH + 0.4 = pD) for a final concentration of ~10 wt.%. A volume of 3 μL was placed in the gasket of the diamond anvil HP FTIR cell, and pressurized with an automated N₂ gas system, as previously described [82]. Pressure calibration was based on the pressure dependence of the stretching vibration of SO₄²⁻ band of BaSO₄ at 983.6 cm⁻¹. Data analysis was performed using Thermo Grams 8.0 software. After buffer subtraction and smoothing of each spectrum, the area of the amide Í band (1700 to 1600 cm⁻¹) was normalized. To identify and determine the number and positions of the sub-bands, Fourier self-deconvolution and 2^nd derivative analysis was used [83]. The amide Í band region of the protein could be decomposed into six sub-bands, and mixed Gaussian-Lorentzian line shape functions were used to fit the peak areas and to determine the relative changes of the secondary structural content [84]. The secondary structure content is in accordance with the data calculated from the crystal structure (1e0s). Small percentage differences in the secondary structure content are likely due to different absorption coefficients of the secondary structure elements.

Contact Maps and Cα-structure based molecular dynamics simulations.

The native contacts in the protein were calculated from the crystal structure (1e0s) using the SMOG server [85]. Using the contact information and the intensity from the HSQC data collected at 1, 1000, 1400 and 1800 bar, the probability of occurrence of all native contacts at each pressure was calculated as previously [44] as the geometric mean of the fractional intensities, $I_i(p)/I_i(\text{atm})$ and $I_j(p)/I_j(\text{atm})$, at pressure, $p$, of the two residues, $i$ and $j$, in the native contact:

$$F{C}_{ij}=\sqrt{\left(\frac{I_i(p)}{I_i(atm)}\right)\times \left(\frac{I_j(p)}{I_{ij}(atm)}\right)}$$ [3]

If one or both residues in a given native contact could not be assigned or tracked with pressure, the fractional value for that contact was set to the average of all experimentally determined fractional contacts. At each pressure, 100 contact lists were created for the Cα coarse-grained simulations in which a given native contact was present across the 100 lists randomly based on its fractional contact at that pressure. Then using these 100 lists, the conformational ensemble was generated at each pressure using the last 50 ns of 100 ns trajectories using Gromacs 4.5.4 [86] as previously described[44]. The representative structures belonging to each basin in the conformational landscape were identified using cluster analysis and converted to all atom representations using Pulchra [87]. Further simulation details can be found in Koduru et al., SI [11]. Cavities in the Arf structure were calculated in Pymol [88], using a detection radius of 3 solvent radii and a detection cutoff of 2 solvent radii.

Ensemble optimization-based modelling (EOM) and DENSS Ab Initio electron density calculations

EOM simulations were conducted using GAJOE (Genetic Algorithm Judging Optimization of Ensemble) [54,55] in the ATSAS suite. The starting ensemble (~10,000 structures) of the EOM calculations were extracted from the trajectories of the above coarse-grained simulations at the indicated pressures and converted to all atom representations using Pulchra [87]. The structural sub-ensemble that best fit the experimental SAXS profiles at the indicated pressures was determined using the χ2. The genetic algorithm was run for over 10,000 generations in each of 200 cycles and a total of 3 independent EOM replicas were performed for both indicated pressures.

Research Highlights.

• Arf6 switching relies on a molten globule ensemble, as we showed for Arf1.

• Although the average global stability of Arf6 is equivalent to that of Arf1, the local distribution of stability is quite different.

• Sequence elements distant from the switch regions that are different in Arf1 and Arf6 are responsible for their specific stability distributions and functional dynamics.

• These sequence-dynamics relationships underlie a heretofore unknown “back-to-front” allosteric mechanism.

• Evolutionary sequence analysis revealed strong non-contact residue covariance in the region containing these allosteric determinants that extends across the ensemble of Ras family of small GTPases.

Significance:

The sequence determinants of protein functional dynamics are central to modulating function for therapeutic and biotechnological applications, yet they remain poorly understood. Here we reveal that, like Arf1, nucleotide switching in Arf6 relies on a functional molten globule state that primes the protein to switch. However, while the average stability of Arf6 is equivalent to that of Arf1, the local stability distribution is distinct. We find that variable sequences on the back side of the in Arf GTPases, distant from the switch regions, impose back-to-front allosteric controls that lead to distinct functional dynamics among homologs. Evolutionary covariance analysis suggests that this back-to-front control of functional dynamics extends across the Ras small GTPase superfamily.

Data availability

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. The original NMR, SAXS, fluorescence and simulation data can be provided by C.A. Royer pending scientific review and a completed material transfer agreement. Likewise our custom Matlab scripts for intensity analysis and fractional contact map calculations can be provided upon request. Requests for the data or scripts should be submitted to: royerc@rpi.edu.