Insight into the Role of Dynamics in the Conformational Switch of the Small GTP-binding Protein Arf1*

Activation of the small GTP-binding protein Arf1, a major regulator of cellular traffic, follows an ordered sequence of structural events, which have been pictured by crystallographic snapshots. Combined with biochemical analysis, these data lead to a model of Arf1 activation, in which opening of its N-terminal helix first translocates Arf1-GDP to membranes, where it is then secured by a register shift of the interswitch β-strands, before GDP is eventually exchanged for GTP. However, how Arf1 rearranges its central β-sheet, an event that involves the loss and re-formation of H-bonds deep within the protein core, is not explained by available structural data. Here, we used Δ17Arf1, in which the N-terminal helix has been deleted, to address this issue by NMR structural and dynamics analysis. We first completed the assignment of Δ17Arf1 bound to GDP, GTP, and GTPγS and established that NMR data are fully consistent with the crystal structures of Arf1-GDP and Δ17Arf1-GTP. Our assignments allowed us to analyze the kinetics of both protein conformational transitions and nucleotide exchange by real-time NMR. Analysis of the dynamics over a very large range of timescale by 15N relaxation, CPMG relaxation dispersion and H/D exchange reveals that while Δ17Arf1-GTP and full-length Arf1-GDP dynamics is restricted to localized fast motions, Δ17Arf1-GDP features unique intermediate and slow motions in the interswitch region. Altogether, the NMR data bring insight into how that membrane-bound Arf1-GDP, which is mimicked by the truncation of the N-terminal helix, acquires internal motions that enable the toggle of the interswitch.

The interplay between the structure of proteins and the nature and timescales of their internal motions, which have become accessible to analysis with recent advances in nuclear magnetic resonance (NMR) spectroscopy methods (1,2), is increasingly recognized as a major component underlying protein functions (reviewed in Refs. 3,4). In particular, the characterization of conformational fluctuations is gaining considerable interest for proteins that undergo large and/or allosteric conformational changes to carry out their functions (5,6). Small GTP-binding proteins (GTPases) 4 represent a fascinating family in that regard, as they undergo large amplitude conformational changes upon converting from their inactive, GDP-bound form to their active, GTP-bound form and adapt their structures to their interactions with multiple regulators and effectors (reviewed in Ref. 7). 15

N relaxation and/or H/D exchange NMR experiments conducted on Rho and Ras family
GTPases have begun to unravel the contribution of internal dynamics to GTPase-based processes, and how the dynamics profiles vary with the nucleotides in presence or the introduction of mutations (8 -10). 15 N-HSQC-based real-time NMR analysis of spontaneous and guanine nucleotide exchange factor (GEF)-stimulated nucleotide exchange (11), as well as spontaneous and GTPase-activating protein (GAP)-stimulated GTP hydrolysis (12) has also been recently established. In this work, we investigate the relationship between structure and dynamics of Arf GTPases, which are central regulators of most aspects of intracellular traffic and its cross-talk to cytoskeleton dynamics (reviewed in Ref. 13). The GDP/GTP switch of Arf proteins represents an extreme case in the GTPase kingdom, characterized by a specific structural device that allows nucleotide exchange to be coupled to membrane recruitment (reviewed in Ref. 14). This device is comprised of the classical switch 1 and switch 2 found in all GTPases and two regions that are unique to Arf proteins: an amphipathic, myristoylated N-terminal helix, and 2 ␤-strands that connect switch 1 to switch 2 (the interswitch) (Fig. 1A). The sequence of conformational changes that take place in the course of Arf activation have been captured by crystallographic snapshots of Arf1, the major eukaryotic isoform, either alone or in complex with GEFs. Structural and biochemical studies have been integrated into a robust model for GEF-stimulated Arf1 activation in cells as follows: (i) the N-terminal helix of Arf1-GDP blocks the interswitch in an eclipsed conformation, which is not competent for binding to membranes (15)(16)(17); (ii) the N-terminal helix opens up and binds to membranes, releasing its hasp on the interswitch (18); (iii) binding of the GEF displaces the switch 1 and promotes a 2-residue register shift of the interswitch, which secures Arf1-GDP to membranes and primes the nucleotide-binding site for GTP (19); (iv) GDP is expelled from the nucleotide binding site (20,21); (v) binding of GTP reorganizes the switch 1 and switch 2 regions (21,22). Whether spontaneous nucleotide exchange follows the exact same route is currently not known.
A few NMR structural studies have also been reported for Arf1 in solution. Backbone assignments have been published for unmyristoylated full-length human Arf1-GDP (23), for the GDP-bound form of a truncated Arf1 mutant that lacks the N-terminal helix (human ⌬17Arf1) bound to GDP (24), and for myristoylated full-length yeast Arf1-GDP (17,25). A structural model of human ⌬17Arf1-GDP was established by refining the crystal structure of Arf1-GDP against residual dipolar couplings (RDCs), which departed considerably from the available crystal structures (24), while the NMR structure of myristoylated yeast Arf1-GDP was much closer to the crystal structures (17). However, none of these studies investigated the contribution of dynamics to the activation process.
In this work, we used NMR spectroscopy to analyze spontaneous nucleotide exchange in real time and determine the internal dynamics of Arf1-GDP, ⌬17Arf1-GDP, and ⌬17Arf1-GTP. Our results, combined with the biochemical and structural data, bring new insight in the interswitch toggle mechanism.

EXPERIMENTAL PROCEDURES
Protein Preparation-Uniformly 15 N, ( 15 N, 13 C) and ( 13 C, 15 N, 2 H) labeled human Arf1 or ⌬17Arf1 were produced in Escherichia coli in M9 minimal media in milliQ water or heavy water supplemented with 15 NH 4 Cl and [ 12 C/ 13 C] glucose. Arf1 constructs and unlabeled human ARNO (Sec7 domain) were purified to homogeneity as described in Ref. 19. All NMR experiments were done in 50 mM HEPES pH 7.3, 5% D 2 O, 150 mM NaCl. The percentage of isotope-labeling was [U-100% 15 N], [U-98% 13 C; U-98% 15 N], and [U-98% 13 C; U-98% 15 N; U-70% 2 H) as checked by mass spectrometry. ⌬17Arf1 was purified as a mixture of GDP-and GTP-bound forms, and was loaded with either GDP, GTP, or GTP␥S by heating the protein samples (100 -600 M) at 310 K during 30 min in the presence of 10 mM Mg 2ϩ and 10 mM nucleotide. Full-length Arf1 was purified as a GDP-bound form.
NMR Spectroscopy-NMR spectra were recorded on Bruker 600 MHz, 700 MHz, 800 MHz, 900 MHz, or 950 MHz Avance spectrometers equipped with a triple resonance triple axes gradient TXI probe for the 700 MHz and triple resonance z axis gradient TXI cryoprobes for the others. One-dimensional 1 H and 31 P experiments were recorded on a 500 MHz Varian Inova spectrometer equipped with a penta-probe.
Backbone Assignments and Structural Analysis of Chemical Shifts-Backbone resonance assignments were done at 298 K using triple resonance experiments (HNCA, HNCACB, CBCA(CO)NH with TROSY versions for deuterated samples). Complete assignments could not be achieved from deuterated samples alone due to a lack of proton back-exchange, and requested the combination of data obtained on protonated and deuterated proteins. 1 H chemical shifts were referenced to DSS, indirect referencing was used for 15 N and 13 C chemical shifts. Direct assignment of ⌬17Arf1-GTP was not possible because of the hydrolysis of GTP in the course of the spectra recording, yielding a mixture of ⌬17Arf1-GDP and ⌬17Arf1-GTP resonances. ( 15 N, 13 C, 1 H) assignments of ⌬17Arf1-GDP and ⌬17Arf1-GTP␥S were thus carried out first, from which the ( 1 H, 15 N) backbone assignments of ⌬17Arf1-GTP could be deduced. Data were processed using NMRpipe (26) and analyzed with Sparky (T. D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco). Secondary structures were predicted from the CЈ, CA, CB, N, and H N chemical shifts by TALOS (27). Normalized chemical shifts variations between forms A and B were computed from 15 N and 1 H N chemical shifts according to Equation 1.
Amide protons in close proximity were identified from threedimensional ( 1 H, 15 N, 1 H) NOESY-HSQC experiments recorded at 800 and 900 MHz on both protonated and deuterated samples using two mixing times of 80 and 120 ms.
Real-time NMR Analysis of Nucleotide Exchange-All experiments were done at a protein concentration of 100 M. GTP␥S (20-fold excess) was added at the beginning of each experiment. The purity of the GTP␥S stock solution (10 mM) was assessed from 31 P one-dimensional spectra. No 31 P signal other than the three characteristic peaks of GTP␥S was detected after 6 h of accumulation, thus demonstrating purity higher than 95% (supplemental Fig. S6). Series of one-dimensional 1 H spectra were recorded at 500 MHz until complete nucleotide exchange. 100 successive ( 1 H, 15 N) HSQC experiments lasting 16 min each were recorded at 800 MHz until complete nucleotide exchange. The temperature (298 K) was calibrated for each spectrometer using the 100% methanol standard Bruker temperature calibration sample. The decrease of peak intensities as a function of time was fitted to a monoexponential function using either Kaleidagraph (one-dimensional experiments) or MATLAB (HSQC). Uncertainties were estimated from 500 simulated data sets using a Monte-Carlo procedure.
A unique protein preparation was split in two identical samples for the one-and two-dimensional experiments. One peak corresponding to the amide proton of Leu-25 in the ⌬17Arf1GDP form was isolated in the one-dimensional spectra, so that its intensity decrease upon GTP␥S exchange could be monitored in both one-and two-dimensional series. This peak was thus used as a reference to check the consistency between the one-and two-dimensional real time experiments. Identical half times were obtained from the analysis of the onedimensional peak and the two-dimensional cross-peak of Leu-25. 15 N Relaxation Experiments-15 N R 1 , 15 N R 2 , and ( 1 H3 15 N) nOe experiments were recorded at 700 MHz. 10 relaxation delays were measured between 30 and 3000 ms for 15 N R 1 and between 0 and 256 ms for 15 N R 2 , with one delay repeated 3 times for error evaluation. Either pseudo-three-dimensional experiments or series of two-dimensional experiments were used. In the pseudo-three-dimensional experiments, the relaxation delay pseudo third dimension was built the fastest and was incremented after each scan. The use of this strategy for ⌬17Arf1-GTP minimized the contribution of the GDP-bound form signal arising from GTP hydrolysis. For heteronuclear ( 1 H3 15 N) nOe experiments, interleaved saturated and unsaturated experiments were acquired. The delay for proton saturation was set to 4 s. All data were processed with nmrPipe (26), intensities were calculated with nmrView (One Moon Scientific Inc). R 1 and R 2 relaxation rates were determined by fitting peak intensities to a single-exponential decay using house-written MATLAB procedures. Uncertainties were estimated as above. The ( 1 H N 3 15 N) nOe values were taken as the ratio between the intensities of corresponding peaks in the spectra recorded with and without saturation of the amide protons. A model-free analysis of relaxation data were performed with the software TENSOR 2 (28).
Relaxation Dispersion Experiments-Spectra were recorded at 700 MHz and 950 MHz at 298K using a pseudo-three-dimensional (interleaved) constant time CPMG sequence optimized as described by Hansen et al. (29,30). The CPMG delay (T CPMG ) was set to 20 ms. 15-17 experiments were acquired with 15 N 180°pulses repetition frequencies ( CP ) between 25 and 1000 Hz during T CPMG . Peak intensities were converted to relaxation rates, and uncertainties in relaxation rates were calculated from repeated experiments as described in Ref. 31. The dispersion curves at the two fields were fitted simultaneously to a global two-state fast exchange (Meiboom equation) using house written MATLAB procedures. Uncertainties were estimated from 1000 simulated data sets using a Monte-Carlo procedure.
H/D Exchange Experiments-All U-[ 15 N] protein samples were lyophilized and dissolved in D 2 O immediately prior to the ( 1 H, 15 N) HSQC experiments. Experiments were carried out at 298K at 700 MHz (Arf1-GDP) or 600 MHz (⌬17Arf1-GDP). The first spectrum was completed ϳ20 min after dissolving the protein in D 2 O. Series of ( 1 H-15 N) HSQC spectra were recorded over 48 h. Data were analyzed using similar procedures as those described for relaxation experiments. Protection factors were obtained as the ratio between the experimentally derived exchange rate constants k ex and the intrinsic exchange rate constants k int calculated with the Sphere software.

Assignments and Cross-validation of NMR and Crystallographic ⌬17Arf1
Data-To enable the real-time and dynamics NMR analysis of Arf1 activation on a sound basis, we established the assignment of human ⌬17Arf1 bound to GDP, GTP, and GTP␥S (supplemental Fig. S1). We first re-assessed the 1 H, 15 N, and 13 C assignments of ⌬17Arf1-GDP, which confirmed and completed the corrections, published by Viaud et al. (32), and assigned the three-dimensional spectrum of ⌬17Arf1 bound to the non-hydrolyzable GTP analog GTP␥S (supplemental Tables S1 and S2). We obtained high quality spectra for ⌬17Arf1-GTP␥S that showed no indication of hydrolysis over time, from which most backbone resonances could be assigned.
⌬17Arf1-GDP and ⌬17Arf1-GTP␥S assignments were eventually used to assign the ( 1 H, 15 N) resonances of ⌬17Arf1-GTP from mixed ⌬17Arf1-GDP/⌬17Arf1-GTP HSQC spectra. GTP␥S induced only 3 significant perturbations as compared with the ⌬17Arf1-GTP spectrum, which were located in the vicinity of the GTP␥S sulfur (supplemental Fig. S1B). This indicates that GTP␥S does not induce notable conformational changes, in agreement with 31 P NMR spectroscopy comparison of Ras bound to GTP and GTP analogs, which identified GTP␥S as the analog that is most similar to physiological GTP (33).
Next, we assessed the consistency between NMR data and structural models obtained from either crystallographic analysis of full-length Arf1-GDP (PDB entry 1HUR, Arf1-GDP Xtal hereafter (15,23)) or NMR analysis using RDCs (PDB entry 1U81, first structure referred to as ⌬17Arf1-GDP RDC hereafter, (24)), the latter being considerably distorted with respect to known structures of Arf and other GTPases. To that purpose we analyzed normalized chemical shift variations between the different forms (CSVs), chemical shifts based secondary structure predictions, and non-sequential H N -H N nOe correlations with respect to each structural model.
CSVs between ⌬17Arf1-GDP and Arf1-GDP are located mainly in switch 1, interswitch and in the C-terminal helix, while none are found in the nucleotide-binding site, at odds with its distorted conformation in ⌬17Arf1-GDP RDC . Their moderate values (Ͻ0.6 ppm) are indicative of at most local and minor conformational differences (supplemental Fig. S2A). By comparison, CSVs between ⌬17Arf1-GDP and ⌬17Arf1-GTP amount to up to 2.5 ppm (supplemental Fig. S2B). Interestingly, chemical shifts back-calculated for Arf1-GDP using Arf1-GDP Xtal and for ⌬17Arf1-GDP using Arf1-GDP Xtal truncated of the N-terminal helix correctly predicted the location of sizeable experimental CSVs (34), indicating that the deletion of the N-terminal helix is sufficient to yield the observed CSVs locations without major conformational changes (supplemental Fig. S3). Consistently, secondary structures predicted from the experimental chemical shifts match the secondary structures measured from Arf1-GDP Xtal significantly better than those calculated from ⌬17Arf1-GDP RDC (100 and 80% of ␣-helices, 80 and 45% of ␤-sheets, respectively). Finally, 135 1 H N -1 H N nOes were extracted from the three-dimensional ( 1 H-15 N-1 H N ) NOESY-HSQC spectra of ⌬17Arf1-GDP. All the corresponding H N -H N distances calculated from Arf1-GDP Xtal were shorter than 5 Å, whereas 44 exceeded 5 Å in ⌬17Arf1-GDP RDC , out of which 11 were larger by up to 4.8 Å (Fig. 1B).
We also assessed that the crystal structure of the ⌬17Arf1-GTP (22) was consistent with the NMR data, including H N -H N distances calculated from nOe correlations for ⌬17Arf1-GTP␥S. Altogether, this analysis establishes that the deletion of the N-terminal helix in Arf1-GDP does not alter its major conformation in solution, and that the crystal structures of Arf1-GDP and ⌬17Arf1-GTP, but not ⌬17Arf1-GDP RDC , are sensible three-dimensional models with which to analyze the structure and dynamics of the GDP/GTP cycle of ⌬17Arf1 in solution.
Real-time NMR Study of the Nucleotide Exchange Process-The HSQC spectra of ⌬17Arf1-GDP and ⌬17Arf1-GTP␥S and those of ⌬17Arf1-GTP and ⌬17Arf1-GTP␥S have 29 and 3 The Conformational Switch of Arf1 DECEMBER 3, 2010 • VOLUME 285 • NUMBER 49 JOURNAL OF BIOLOGICAL CHEMISTRY 37989 non-overlapping crosspeaks, respectively. The nucleotide exchange process can thus be followed in the same sample from the relative intensities of each set of crosspeaks ( Fig. 2A). We also used the fact that ⌬17Arf1 is purified from E. coli bound to 15 N-labeled nucleotides yielding a mixture of GTP-and GDPbound forms from which nucleotide dissociation upon addition of unlabeled GTP␥S can be followed in one-dimensional proton spectra (Fig. 2B). 1 H one-dimensional and ( 1 H, 15 N) two-dimensional spectra were collected over time after addition of 20-fold excess GTP␥S until complete exchange of GDP-and GTP-for GTP␥S (Fig. 2, A and B). We could therefore monitor simultaneously the kinetics of GDP and GTP dissociation (k off_GDP and k off_GTP , Fig. 2, B and C) and of ⌬17Arf1-GDP/ GTP environmental changes (k 1_GDP and k 1_GTP ). k off_GDP and k off_GTP indicate that GTP␥S replaces GDP slightly faster (k off_GDP ϭ 0.30 Ϯ 0.03 h Ϫ1 ) than it does for GTP (k off_GTP ϭ 0.12 Ϯ 0.02 h Ϫ1 ). The half times of decay for ( 1 H N,15 N) peaks of ⌬17Arf1GDP were similar for most residues (Fig. 2D and supplemental Fig. S4), indicating a global process with k 1_GDP ϭ 0.15 Ϯ 0.02 h Ϫ1 . A slightly smaller k 1_GTP value of 0.057 Ϯ 0.002 h Ϫ1 was obtained from the subset of GTP-associated crosspeaks that vary upon GTP␥S exchange. Surprisingly, the rates of environmental changes k 1_GDP and k 1_GTP were about twice smaller than the off rates of nucleotides. The relevance of this difference is assessed by the measurement of an identical decrease rate of the amide proton one-dimensional peak and two-dimensional ( 1 H, 15 N) crosspeak of Leu-25 in the GDP state (Fig. 2C).
We then analyzed nucleotide exchange stimulated by the Sec7 catalytic domain of the GEF ARNO. In the case of ⌬17Arf1, addition of 1% molar equivalent of ARNO made the kinetics of the nucleotide exchange too fast to be monitored even by one-dimensional 1 H experiments. In contrast, the ( 1 H-15 N) HSQC protein chemical shifts of full-length Arf1-GDP were unaffected by the addition of GTP or GTP␥S, regardless of the presence of ARNO, as expected from previous biochemical work (reviewed in Ref. 35). Surprisingly, the bound 15 N-labeled GDP was however replaced by unlabeled GTP␥S in the presence of ARNO, as shown by the disappearance of the 15 N doublet and the appearance of a 14 N singlet. Contamination of the GTP␥S sample was ruled out by 31 P spectroscopy. The fact that the additional phosphate of GTP␥S is accommodated in the GDP conformation of Arf1 without yielding protein CSVs in the nucleotide binding site suggests that the additional group is expelled from the nucleotide binding site.
Globally, these results suggest the presence of intermediates of the spontaneous exchange reaction that are either empty or can accommodate GDP or GTP without significant conformational change. These states could resemble those evidenced by x-ray crystallography of GDP-bound intermediates of ⌬17Arf1-GDP bound to ARNO (20). In the case of Arf1-GDP, the presence of ARNO enhances some local plasticity, sufficient to release the nucleotide and accommodate a GTP, but not to produce the conformational switch.

⌬17Arf1-GDP Exhibits Dynamics over a Larger Range of Timescales than Arf1-GDP and ⌬17Arf1-GTP
Whereas the structural components of the GDP/GTP exchange of Arf1 are now well established, the contribution of internal dynamics has not been investigated. Dynamics can be analyzed over a large range of time scales by 15    GDP, and ⌬17Arf1-GDP dynamics by 15 N relaxation and CPMG 15 N RD and, for Arf1-GDP and ⌬17Arf1-GDP by native-state H/D exchange. 15 N relaxation rate constants R1, R2, and nOe, 15 N CPMG RD and H/D exchange data are given in supplemental Fig. S5 and Tables S3 and S4, respectively. 15 N relaxation rate constants could be measured for most residues in all Arf1 forms, except in the switch 2 due to either broad (⌬17Arf1-GDP and Arf1-GDP) or overlapping (⌬17Arf1-GTP) resonances. They were analyzed with a model-free formalism (36). 15 N relaxation data indicate that all Arf forms have an isotropic global rotational motion (overall global correlation time c at 298K for Arf1-GDP, ⌬17Arf1-GDP, and ⌬17Arf1-GTP equal to 10.87 Ϯ 0.03, 10.30 Ϯ 0.04, and 10.67 Ϯ 0.05 ns, respectively) and behave as globular monomers in solution.
Comparison of internal dynamics shows that ⌬17Arf1-GTP is the least dynamic form, displaying only highly restricted internal motions. This is apparent from the small number of residues, all of which are located near the switch 2, that have generalized order parameter S 2 values smaller than 0.85, inter-mediate internal correlation times ( i ) in the nanosecond range and conformational exchange contributions (R ex ) associated to motions in micro-millisecond range (Fig. 3). Consistently, RD curves were flat for most residues, suggesting that the protein backbones have mainly localized motions in the millisecond regime (Fig. 4, right).
The dynamics of Arf1-GDP is somewhat more complex. Although most residues have S 2 values larger than 0.85, several exhibit internal correlation time i in the nanosecond range. This is notably the case for the N-terminal helix and switch 1 regions. R ex contributions are predicted for these regions, although they are only moderately predictive of actual millisecond exchange processes considering their rather small values (5 Hz range) (Fig. 3). Accordingly, CPMG RD experiments indicate the existence of motions in the micro/millisecond timescale for only a small number of residues, in particular in the N-terminal helix (Fig. 4, left). Finally, nativestate H/D exchange experiments identified amide protons that were exchanged within minutes to hours, and were located across the entire structure (Fig. 6A, top). In particular, all amide protons located in the N-terminal helix and in the switch 1 are solvent-exchanged within minutes, suggesting that these structures are flexible in this form.
The dynamics of ⌬17Arf1-GDP departed significantly from those of both ⌬17Arf1-GTP and Arf1-GDP. While most residues have S 2 values larger than 0.85 and only a small number have intermediate internal correlation times i in the ns range, R ex contributions in the switch regions are significantly larger than those calculated for Arf1-GDP (Fig. 3). In contrast with Arf1-GDP and ⌬17Arf1-GTP, a large number of residues exhibit non-flat CPMG RD curves, which reveals extensive internal motions in the micro-millisecond time scale. Interestingly, a set of residues exhibiting significant RD was located along the whole ␤-sheet (Fig. 4, middle panel). The exchange rate constant k ex equal to 1530 Ϯ 110 s Ϫ1 obtained from the global fit of the data at two fields is compatible with a two-state fast exchange (Fig. 5). This fast exchange regime is corroborated by the quadratic dependence of the exchange contributions R ex with the magnetic field (ratio of R ex at 22.3T and 16.45T equal to 1.84 for all residues). Under this condition, the fitted parameter is ⌬␦ app ϭ [p B (1 Ϫ p B )] (1/2) ⌬␦ and it is not possible to extract the population of each state and the difference in chemical shifts ⌬␦ between the states for each residue. However, it should be noted that because p B is the same for all residues (global exchange), the relative variations of ⌬␦ between residues along the sequence is independent of the population. The results are given in supplemental Table S3.   DECEMBER 3, 2010 • VOLUME 285 • NUMBER 49 Native-state H/D exchange experiments identified essentially the same amide proton in slow exchange as in Arf1-GDP, except for a striking decrease of the protection factors in the vicinity of the interswitch region (Fig. 6A, bottom). This decrease of the protection factors in ⌬17Arf1-GDP as compared with Arf1-GDP indicates that 8 of the hydrogen bonds that are predicted from the crystal structure are likely to be less stable in ⌬17Arf1-GDP (Fig. 6B, right) than in Arf1-GDP (Fig.  6B, left, supplemental Table S4). Notably, all hydrogen bonds between the ␤-strands of the interswitch (␤2-␤3), and additional hydrogen bonds between the interswitch and strand ␤1 or switch 1 are less stable in ⌬17Arf1-GDP than in Arf1-GDP. As a conclusion, our results show that buried interswitch residues that are protected in Arf1-GDP become unprotected in ⌬17Arf1-GDP. Thus the truncation of the N-terminal helix induces motions that allow H/D exchange to take place deep in the protein, involving the disruption of internal hydrogen bonds (Fig. 7).

DISCUSSION
To establish the first dynamics analysis of an Arf GTPase on a sound basis, we first completed the assignment of human ⌬17Arf1 bound to GDP, GTP, and GTP␥S, and carefully assessed that available NMR information for Arf1 and ⌬17Arf1 was fully consistent with crystallographic three-dimensional data. This ruled out the conclusion of a previous work (24), which inferred that the crystal structures were not relevant to the structures in solution. This also shows that RDCs should be used with caution in structure calculations, as they also incorporate internal dynamics contributions, which as we show here are sizable in ⌬17Arf1-GDP.
NMR relaxation dispersion techniques recently emerged as a powerful approach to analyze the propagation of dynamics across proteins on a per-residue basis in signaling or enzymatic processes (reviewed in Refs. 3,4). Our ensemble of Arf1 assignments allowed us to undertake an extensive analysis of the dynamics profiles of Arf1-GDP, ⌬17Arf1-GDP, and ⌬17Arf1-GTP in time scales ranging from fast and intermediate to slow motions. Notably, incorporation of native state H/D exchange data allowed us to gain access to motions in the minutes to hours range. Our analysis reveals that ⌬17Arf1-GTP and Arf1-GDP dynamics are restricted to fast and local motions, whereas ⌬17Arf1-GDP displayed a complex dynamics behavior with components in the fast, intermediate and slow regimes. Consistent with these observations, ⌬17Arf1-GDP is thermodynamically less stable than ⌬17Arf1-GTP and Arf1-GDP, as reflected by their denaturation temperatures measured by the thermo-fluorescence assay (respectively Tm ϭ 63.4, 68.4, and 73.6°C). The switch 1 and N-terminal helix of Arf1-GDP displayed however intrinsic dynamics, consistent with the fact that these elements are displaced first in the exchange reaction (18,19,21). These internal motions may therefore prime Arf1-GDP for the initiating event of the exchange reaction.
⌬17Arf1-GDP can be considered as a mimic of membraneassociated Arf1-GDP, in which the truncation of its N-terminal helix is equivalent to the displacement of this helix as it binds to membranes in full-length myristoylated Arf1-GDP. This truncation unlocks the retracted conformation of the interswitch, making Arf1-GDP competent for the subsequent membranesecuring step (19). Accordingly, ⌬17Arf1-GDP is readily activated by GDP/GTP exchange in solution, whereas full-length Arf1 cannot be activated under these conditions (reviewed in Ref. 35). However, how Arf1-GDP accommodates the toggle of ∆ FIGURE 6. Native-state H/D exchange experiments. A, logarithm of solvent protection factors (log(PF HD )) of the amide protons of Arf1-GDP (black) and ⌬17Arf1-GDP (red) as a function of the protein sequence. Zero values correspond to residues whose amide protons are already fully exchanged in the first HSQC spectrum. Values higher than six correspond to residues that were not exchanged after 120 h. B, H/D exchange in the interswitch region and flanking ␤1 and switch 1 ␤-strand in Arf1-GDP (left) and ⌬17Arf1-GDP (right). Residues whose amide protons exchange with deuterium faster than minutes, in the order of hours, and slower than a week are represented in pink, purple, and blue, respectively. Amide protons of residues that cannot be assigned are shown in gray (ND). Hydrogen bonds are highlighted in pink for those involving fast exchanged amide protons (Ͻmin), in gray for ND, and in black otherwise.  DECEMBER 3, 2010 • VOLUME 285 • NUMBER 49 the interswitch in the core of its central ␤-sheet while maintaining its folded structure has remained unexplained. The comparative analysis of internal dynamics between ⌬17Arf1-GDP, ⌬17Arf1-GTP, and Arf1-GDP provides for the first time some insight into this central event of Arf and Arf-related GTPases activation (reviewed in Ref. 14).

The Conformational Switch of Arf1
Altogether, our combined structural and dynamics NMR analysis of Arf1 provides novel insight into the allosteric propagation of information between the side of Arf1 that faces the membrane, and its opposite face that binds nucleotides and cellular partners. The localization of the millisecond range motions along the ␤-sheet down to the switch regions and the nucleotide-binding site reveals how the release of the N-terminal helix opens a "front-back" communication path across the protein. This communication path is blocked in the full-length protein. We can thus speculate that the combination of interstrand motions evidenced by CPMG and H/D exchange experiments in the ms to seconds timescales altogether converge toward the opening of the interswitch domain through a lateral process that involve a melting/dissociation of the central ␤-strands. Inter-strand motions in the ms range trigger higher amplitude motions in the interswitch and switch 1 domains, that may finally drive the whole conformational change from the GDP-to the GTP-bound states orders of magnitude slowly (21,22). The transient formation of more opened states shown through H/D exchange thus possibly favors the stabilization of the GTP form (Fig. 7).
Our results indicate how, in membrane-associated Arf1-GDP, the displacement of the N-terminal helix through its membrane anchoring renders Arf1-GDP "softer" in the vicinity of the interswitch, and reveal conformational sub-states in exchange at the millisecond timescale that prime the subsequent toggle of the interswitch. The interswitch may then use these acquired internal motions to toggle without disrupting the overall structure of Arf1.