Protein Engineering Design and Selection

PEDS Advance Access originally published online on June 26, 2008
Protein Engineering Design and Selection 2008 21(9):577-587; doi:10.1093/protein/gzn035

This Article Abstract FREE Full Text (PDF) Supplementary Data All Versions of this Article: 21/9/577    most recent gzn035v1 Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Crowhurst, K. A. Articles by Mayo, S. L. Search for Related Content PubMed PubMed Citation Articles by Crowhurst, K. A. Articles by Mayo, S. L. Social Bookmarking          What's this?

© The Author 2008. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org

NMR-detected conformational exchange observed in a computationally designed variant of protein Gβ1

Karin A. Crowhurst^1,3 and Stephen L. Mayo^1,2,4

¹Division of Biology ²Division of Chemistry and Chemical Engineering, California Institute of Technology, MC 114-96, Pasadena, CA 91125, USA

⁴ To whom correspondence should be addressed. E-mail: steve{at}mayo.caltech.edu (S.L.M.); karin.crowhurst{at}csun.edu (K.A.C.)

	Abstract

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Detailed biophysical characterization of computationally designed proteins has become increasingly important in order to thoroughly understand the properties of these variants compared with wild-type and to apply this knowledge to future designs. The protein dynamics and structural properties of a computationally designed variant (1.5) of the β1 domain of streptococcal protein G (Gβ1) were measured using multinuclear NMR methods. Results from relaxation, diffusion and hydrogen exchange experiments indicate that the variant weakly self-associates at NMR concentrations, with evidence for multiple binding sites. Although comparison of fast (ps–ns) timescale motions shows only small differences in dynamics between 1.5 and wild-type, results from the measurement of intermediate (µs–ms) timescale motions are very different. Significant backbone conformational exchange has been observed in the variant at positions all along the sequence, whereas the wild-type Gβ1 shows little evidence for this type of motion. This increased conformational exchange in 1.5 has been attributed to core overpacking resulting from the incorporation of two large hydrophobic side chains and the loss of an aromatic T-stacking interaction. These data highlight, in detail, the potential consequences of incorporating major perturbations in the core of a protein and the need to carry out more detailed analyses of the biophysical properties of designed proteins in order to better understand and predict the effects of mutations.

Keywords: backbone dynamics/computational protein design/conformational exchange/ORBIT/protein Gβ1

	Introduction

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Computational protein design has progressed significantly over the past 10 years, moving from simple proof-of-principle designs to full-sequence and de novo fold designs (Dahiyat and Mayo, 1997; Harbury et al., 1998; Kuhlman et al., 2003; Looger et al., 2003). Most research has involved efforts to learn how to generate protein sequences that fold to specified templates and remain stably folded. For the past several years, researchers have started tackling the design of proteins that carry out biological functions. Despite this, the standard method of characterizing new sequences has remained largely the same. In general, the success of each design continues to be judged based primarily on the degree of its thermal stability and how closely its NMR or X-ray crystal structure matches the target fold. However, a great deal of theory and knowledge will be required to achieve the transition from the design of proteins with specific folds to those with specific biological functions. As a result, increasing numbers of protein design researchers are realizing that the pursuit of more detailed biophysical characterization of their designs is essential to bridge the gap (Gillespie et al., 2003). Considering the emphasis on the importance of core packing in computational design software development over the past decade (Johnson et al., 1999; Walsh et al., 2001; Ventura et al., 2002), researchers are also beginning to understand that analysis of protein dynamics in addition to protein structure could be a critical component to future design capabilities. Detailed biophysical analysis can help to explain the properties (such as thermal stability) of previously designed variants and to aid in improving the rate of success in sequence generation and in the incorporation of desired functionality.

Interestingly, the few currently existing publications that describe protein motions in designed proteins have already provided a wide variety of results. On one end of the range, the de novo designed protein ₃D has a core that is much more ‘malleable’ than naturally occurring proteins of a similar size (Walsh et al., 2001). On the other end of the range, both the redesigned human U1A protein (Dobson et al., 2006) and the 7-fold core mutant of ubiquitin (Johnson et al., 1999) exhibit fast timescale dynamics that are virtually indistinguishable from that of their respective wild-type proteins, but have little or none of the observed wild-type conformational exchange. The limited number of reports on dynamics in designed proteins prevents us from drawing general conclusions about the causes and functional implications of increased or decreased protein motions, but the reports that do exist suggest that the characterization of protein motions in these designs can provide significant insight into the factors governing design success.

The incorporation of algorithms allowing for backbone positions that differ from the starting conformation (in addition to side chain placement) has also been a long-standing goal of many computational design groups, since it permits the generation of a greater variety of sequences and is likely to be an important component of enzyme design. Although some steps have been taken towards full backbone variability (Harbury et al., 1998; Desjarlais and Handel, 1999; Larson et al., 2002; Zavodszky et al., 2004; McCammon, 2005), success is still elusive. Previously, we used the β1 domain of streptococcal protein G (Gβ1, Fig. 1A) as a template to test the robustness of the ORBIT protein design software package in choosing amino acids to respond to changes in the positioning of supersecondary structural elements (Su and Mayo, 1997). The overall goal was to develop methods that would permit small rearrangements in backbone conformation during sequence design calculations. One design from this exercise, originally named h_1.0[+1.50 Å] and now referred to as 1.5, is a variant with six mutations at the most buried (core) positions (Fig. 1B). It was generated while attempting to obtain sequences that could support a Gβ1 backbone conformation in which the helix was translated by 1.5 Å away from the plane of the β-sheet (Su and Mayo, 1997). The design was observed to fold to a stable tertiary structure with a melting temperature, T_m = 73°C, which is only 12°C lower than wild-type.

View larger version (41K): [in this window] [in a new window] [Download PowerPoint slide]   Fig. 1. (A) Labeled ribbon diagram of protein Gβ1 (PDB ID: 1PGA) and (B) sequence alignment of wild-type (WT) Gβ1 with variant 1.5. The locations of the six mutations in 1.5 are indicated in darker, bold type. The figure was generated using Pymol (DeLano, 2002).

 
When the NMR solution structure of 1.5 was solved (Ross et al., 2001), two interesting observations were made. The first was that the fold more closely matched the wild-type structure than the template used in the computational design process, most likely rendering the core overpacked (Ross et al., 2001). Second, evidence for dynamic behavior in the backbone and side chains of the variant was observed. In Ross et al. (Ross et al., 2001), it is hypothesized that these two observations are connected and that, in order for the protein to remain folded, the overpacking of the core must be offset by something such as local structural fluctuations.

In this report, we use a variety of solution-state NMR methods to pursue a detailed analysis of the motions of backbone amide groups at both fast (ps–ns) and intermediate (µs–ms) timescales to assess the impact of the core mutations and unexpected structural properties of the computationally designed variant 1.5. We have found evidence of transient dimer formation in the protein at NMR concentrations and extensive intermediate timescale conformational exchange that is not present in the wild-type protein. When probing fast timescale backbone motion, however, we observe little difference between the wild-type and variant, except for slightly increased rigidity in the variant. Results indicate that the causes of the increased conformational dynamics arise primarily from core overpacking and the loss of an aromatic T-stacking interaction. Also discussed is the correlation between protein dynamics and core overpacking, the correlation between dynamics and thermal stability, attempts to map the dimer interfaces, as well as the implications of the results to protein design.

	Materials and methods

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Sample preparation

All isotopically labeled reagents were purchased from Cambridge Isotope Labs.

The variant construct has been described previously (Su and Mayo, 1997). Uniformly ¹⁵N-, ¹³C/¹⁵N- and ¹⁵N/²H-labeled proteins were expressed in Escherichia coli BL21(DE3) cells transformed with a pET-11a vector containing the appropriate gene (for wild-type Gβ1 or 1.5).

For undeuterated samples, a single colony of freshly transformed E.coli BL21(DE3) carrying the 1.5 or wild-type Gβ1 plasmid was inoculated into 50 ml M9 minimal medium containing 100 mg l^–1 ampicillin (enriched with either ¹⁵NH₄Cl/¹H,¹²C glucose or ¹⁵NH₄Cl/¹H,¹³C glucose) and shaken at 37°C overnight. This culture was centrifuged (5000 rpm, 30°C, 5 min) and the pellet was resuspended into 950 ml minimal medium containing 100 mg l^–1 ampicillin. The flask was shaken at 37°C until A₆₀₀0.7 at which time the culture was induced by adding IPTG (250 mg l^–1 of culture) and then transferred to 30°C to shake overnight before harvesting.

For deuterated samples, carbenicillin (100 mg l^–1) was utilized in all cultures instead of ampicillin due to its resistance to breakdown during long incubation times. A single colony of freshly transformed E.coli BL21(DE3) carrying the variant Gβ1 plasmid was inoculated into a 100 ml solution of LB and shaken at 37°C ( 6 h). A small fraction of this culture was used to inoculate 100 ml of ¹⁵NH₄Cl/H₂O/¹H,¹²C-glucose M9 medium and the flask was shaken (250 rpm, 37°C) until A₆₀₀0.9, at which time the culture was centrifuged (5000 rpm, 37°C, 5 min). A portion of the pellet was resuspended into 1 l ¹⁵N²H₄Cl/²H₂O/²H,¹²C-glucose M9 medium in a 2.8 l fernbach flask (for a starting A₆₀₀0.1). The solution was shaken (250 rpm, 37°C) until A₆₀₀0.6 at which time protein expression was induced by adding IPTG (250 mg l^–1 of culture); the flask was then transferred to 30°C and grown overnight before harvesting.

Pellets were lysed using a freeze–thaw protocol (Johnson and Hecht, 1994) and purified by reversed-phase high-performance liquid chromatography using an acetonitrile/water gradient containing 0.1% trifluoroacetic acid.

NMR experiments

All NMR experiments were recorded at 25°C on protein samples in sodium phosphate buffer (50 mM, pH 6), 10% D₂O, 0.2 mM DSS (unless otherwise noted), recorded on a Varian Inova 600 MHz spectrometer equipped with a pulsed-field gradient triple resonance probe.

Assignment of chemical shifts  Backbone chemical shift assignments were performed on a 1.5 mM, ¹³C/¹⁵N-labeled sample of 1.5. The ¹⁵N-¹H HSQC (Kay et al., 1992) utilized 128 x 640 complex points in t₁ and t₂, respectively, and spectral widths of 2000.0 and 7509.6 Hz for F₁ and F₂. The ¹³C-¹H HSQC (Santoro and King, 1992) utilized 256 x 469 complex points in t₁ and t₂, respectively, and spectral widths of 12 000.0 and 11 001.1 Hz for F₁ and F₂. The 3D HNCCβ experiment (Muhandiram and Kay, 1994) utilized a matrix of 64 x 32 x 625 complex points and spectral widths of 10 000.0, 2000.0 and 7509.6 Hz for F₁, F₂ and F₃, respectively. Chemical shift assignments for wild-type Gβ1 (1.8 mM, ¹³C/¹⁵N-labeled sample) were obtained elsewhere (D.J. Hodgson and K.A. Crowhurst, unpublished results).

Diffusion experiments  Spectra were recorded on ¹⁵N-labeled samples of 1.5 at concentrations of 0.95, 1.7, 2.0, 2.3 and 2.9 mM and on a 2.5 mM, ¹⁵N-labeled sample of wild-type Gβ1. A 2D version of the stimulated echo LED experiment (Choy et al., 2002) was recorded in series (varying the field strength of the encoding and decoding gradients (g₂)) to measure translational diffusion coefficients. The spectra were acquired with spectral widths of 7509.6 Hz with 1024 complex points in F₂ (¹H) and 2000 Hz with 128 complex points in F₁ (¹⁵N). Values for g₂ were 4.05, 6.08, 8.10, 10.13, 12.15, 14.18, 16.20, 18.23, 20.25, 24.30, 26.33 and 30.38 G/cm (recorded in random order to reduce systematic errors during data measurement). Peak intensities were fitted to the equation:

1

where G is the field strength of the encoding and decoding gradients (g₂), I(G) is the peak intensity at measured at field strength G, I(0) is the peak intensity at G = 0, is the proton gyromagnetic ratio (2.675 x 10⁴ rad G^–1s^–1), D is the diffusion coefficient and the delays ( and ) are defined in the original pulse sequence (Choy et al., 2002). For these experiments, values of = 2.4 ms and = 205 ms were used.

Hydrogen exchange experiments Two types of hydrogen exchange experiments were recorded on both 1.5 and wild-type Gβ1 proteins. Slow exchange of amide protons was monitored by standard hydrogen–deuterium (H/D) exchange experiments, in which lyophilized, protonated samples were dissolved in deuterated buffer (50 mM PO₄ in 100% D₂O) and a series of ¹⁵N-¹H HSQC spectra (40 minutes each) were recorded over 24 h, monitoring the rate of decrease in peak intensity with time. Spectra were recorded with 128 x 640 or 128 x 1024 complex points in t₁ and t₂, respectively, and spectral widths of 2000.0 and 7509.6 Hz for F₁ and F₂ on a ¹³C/¹⁵N-labeled sample of 1.5 (2.2 mM, pH* 6.3) and on ¹⁵N-labeled wild-type Gβ1 (2.6 mM, pH* 5.8). Fast exchange of amide protons with water was monitored using a CLEANEX-PM pulse sequence (Dempsey, 2001) on ¹⁵N-labeled samples of 1.5 (2.6 mM, pH 6) and wild-type Gβ1 (2.5 mM, pH 6). Each 2D spectrum was recorded with spectral widths of 7509.6 Hz with 614 complex points in F₂ (¹H) and 2000 Hz over 128 complex points in F₁ (¹⁵N). A series of 10 spectra were collected with CLEANEX mixing times ranging from 2.5 to 100 ms. A least-squares fitting procedure was employed to extract NH proton exchange rate constants, k_obs (Hwang et al., 1998). Amide proton protection factors were calculated from the results of all hydrogen exchange experiments as k_int/k_obs, where k_int are the intrinsic exchange rate constants that were estimated using the internet program SPHERE (http://www.fccc.edu/research/labs/roder/sphere/) based on methods developed by Englander and co-workers (Bai et al., 1993).

Relaxation experiments  Spectra were recorded on a ¹⁵N/²H-labeled sample of wild-type Gβ1 (1.5 mM) and ¹⁵N/²H-labeled samples of 1.5 (1.1 mM and 3.0 mM). ¹⁵N longitudinal (R₁) and transverse (R₂) relaxation rate constants and ¹⁵N-¹H steady-state NOEs were measured using two-dimensional ¹H-detected heteronuclear pulse sequences employing sensitivity-enhanced gradient coherence selection (Farrow et al., 1994). All spectra were recorded with spectral widths of 7509.6 Hz with 1024 complex points in F₂ (¹H) and 2000 Hz with 128 complex points in F₁ (¹⁵N). R₁ and R₂ spectra were acquired with 16 transients, while the NOE spectra were acquired with 64 transients. R₁ values were determined from 12 spectra with 10 delays of 0.01, 0.05, 0.10 (x2), 0.15, 0.20, 0.25, 0.40, 0.60 (x2), 0.80 and 1.00 s. R₂ values were determined from 11 spectra with nine delays of 0.01, 0.03, 0.05 (x2), 0.07, 0.09, 0.13, 0.17, 0.21 (x2) and 0.25 s. NOE data were obtained by recording one spectrum with a 3 s recycle delay and a 3 s saturation and another (control) spectrum with no saturation and a 6 s recycle delay. The NOE and control spectra were recorded in triplicate in an interleaved manner. Longitudinal (_z) and transverse (_xy) ¹H–¹⁵N dipolar/¹⁵N CSA (chemical shift anisotropy) cross-correlation relaxation rate constants were also measured (Kroenke et al., 1998), with 16 and 64 transients acquired in the I_auto and I_cross experiments, respectively. Spectra were recorded with spectral widths of 7509.6 Hz with 2048 complex points in F₂ (¹H) and 2000 Hz over 200 complex points in F₁ (¹⁵N), with a 2.5 ms interscan delay. The _z values were obtained from spectra with relaxation delays of 150, 225, 300, 375 and 460 ms, and the _xy values were obtained from spectra with relaxation delays of 32, 53.4, 74.8, 96.1 and 106.8 ms.

Data processing and analysis

All NMR spectra were processed and analyzed on a Linux-based PC using NMRPipe/NMRDraw (Delaglio, 1995) and NMRView (Johnson and Blevins, 1997; Johnson and Blevins, 1994) software, respectively. Relaxation data were analyzed using both NMRView (for NOE data) and CurveFit software (Palmer III, 1998). Error estimates for R₁ and R₂ data were calculated using Monte Carlo simulations. Steady-state NOE values were calculated from the ratio of peak intensities from spectra with and without proton saturation, and errors were estimated from the variation in values from triplicate spectra.

Analysis of raw data from the cross-correlation relaxation experiments to obtain the values of _xy and _z was as described in the original publication of the experiments (Kroenke et al., 1998). Estimation of the transverse relaxation rate constant, R₂°, that is free of conformational exchange contributions, is calculated from:

2

where

3

_xy and _z are the transverse and longitudinal cross-correlation rate constants, respectively and _N and _H are the gyromagnetic ratios for ¹⁵N and ¹H, respectively (Kroenke et al., 1998). The value of the conformational exchange contribution can then be calculated for each ¹⁵N spin from:

4

Reduced spectral density analysis was performed using a suite of Mathematica (Wolfram Research, 2005) notebooks written by the Spyracopoulos lab at the University of Alberta for analysis of ¹⁵N backbone relaxation data (Spyracopoulos, 2006). The input parameters for these calculations utilized R₁, R₂° and NOE data and the results should, therefore, reflect only fast timescale motions (with no contribution to J(0) from intermediate timescale motions).

Chemical shift assignments, R₁, R₂, R_ex and hydrogen exchange experimental data have been deposited online at the BMRB (accession number 15408).

	Results

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Chemical shift assignment

Previous experiments on the 1.5 protein variant were carried out using homonuclear NMR techniques (Su and Mayo, 1997; Ross et al., 2001). For the current work, assignment of ¹H and ¹⁵N resonances was required; chemical shift assignments were, therefore, obtained on uniformly ¹³C/¹⁵N-labeled 1.5, utilizing 2D ¹⁵N-¹H HSQC and 3D HNCCβ experiments (Kay et al., 1992; Muhandiram and Kay, 1994).

Observation of a 1.5 homodimer

Significant differences in transverse relaxation rate constant (R₂) values were observed when comparing data from the higher (3.0 mM) and lower (1.1 mM) concentration samples of 1.5 (which will herein be indicated by 1.5(H) and 1.5(L), respectively). Since R₂ is heavily influenced by changes in molecular weight, it became clear that the variant forms a homodimer in solution at NMR concentrations (more details of relaxation experimental results are discussed in later sections). Because the interaction is clearly weak (based on the results of diffusion experiments, below), efforts to identify the binding interface of the homodimer have proven difficult. First, chemical shift analysis indicates only small changes in both ¹H and ¹⁵N chemical shifts (less than 0.04 and 0.15 ppm, respectively, data not shown), even over a large change in concentration (0.1 to 3.8 mM), and do not highlight a clear binding interface. Second, hydrogen exchange experiments show no increases in protection factors in any region of 1.5 compared to wild-type Gβ1 (in fact, there are significant decreases observed: see the hydrogen exchange experimental results, below). Finally, NOE experiments designed to measure intermolecular contacts (Zwahlen et al., 1997) would be unsuccessful on such a weak dimer, since the transient nature of the interaction would preclude observation of any crosspeaks. Based on concentration-dependent changes observed for some residues in several experiments (which include diffusion, hydrogen exchange and relaxation experiments), we are able to speculate on possible dimer interface sites (see Discussion), but detailed structural characterization of the complex is currently not feasible.

Diffusion experiments

The dissociation constant for self-association of 1.5 was expected to be in the low millimolar range; as a result, we opted to use NMR diffusion experiments to estimate K_D. A 2D version of the stimulated echo LED (Longitudinal Encode-Decode) experiment was recorded rather than the standard 1D version, since it removes background ‘noise’ from the peak intensity measurements arising from sample impurities (i.e. small molecules such as glycerol) and reduces the contribution of protein aggregates to the signal (Choy et al., 2002). Raw data were recorded and analyzed as outlined in Materials and Methods.

Translational diffusion coefficients were measured at five different concentrations of 1.5. An estimation of the dissociation constant for formation of the weak homodimer was obtained by fitting the data to a four-parameter sigmoid equation (Fig. 2):

5

where D_o is the lowest diffusion coefficient value, a is the range of D values, b is the width of the transition and x_o, the transition point for the curve, is also equal to the dissociation constant (K_D). The K_D for monomer–dimer exchange in 1.5 can, therefore, be estimated at 1.94 ± 0.02 mM. Considering the weakness of the interaction and taking into account the results of other experiments performed to characterize the dimer (including H/D exchange, see below), it is possible that the self-association is non-specific and that several different dimer orientations are sampled in solution.

View larger version (11K): [in this window] [in a new window] [Download PowerPoint slide]   Fig. 2. Diffusion coefficient as a function of [1.5] monitored using 2D stimulated echo LED NMR experiments. Data are fit using a standard four-parameter sigmoid equation.

 
Hydrogen exchange experiments

Amide proton exchange was monitored for both 1.5 and wild-type Gβ1, utilizing standard H/D exchange methods and the CLEANEX-PM pulse sequence (Hwang et al., 1997) to measure slow and fast timescale exchange, respectively. Per-residue protection factors are displayed in Fig. 3. There is a significant difference between amide protection in the wild-type and variant, particularly in regions including the β₂ strand (residues 12–19), the N-terminal tail of the -helix (residues 21–25) and the long loop between residues 36 and 41. Even the maximum protection factors for residues in 1.5 are at least one order of magnitude lower than that of the wild-type protein; after 24 h of H/D exchange, there was very little peak intensity remaining in the 1.5 spectrum, while there were still a number of fairly intense peaks remaining in the wild-type spectrum.

View larger version (34K): [in this window] [in a new window] [Download PowerPoint slide]   Fig. 3. Log-scale protection factor (for backbone amide proton) as a function of residue number for (A) wild-type Gβ1 and (B) 1.5. The positions of secondary structural elements are indicated above (A). Dark-colored bars represent data acquired through standard H/D exchange experiments (for amide hydrogens that exchange slowly), whereas light-colored bars represent data recorded using CLEANEX-PM experiments (for direct measurement of rapid amide proton exchange with water). Bars marked with a white star (*) indicate the most protected amides; for these, insufficient exchange occurred during the period of the experiment and prevented our ability to fit the data to obtain rates of exchange. The maximum log scale value for wild-type protection factors is 6.5, but the y-axis of A is limited to a maximum of 4.5 in comparison with B. The figure was generated using SigmaPlot (SPSS, 2001).

 
Relaxation measurements

The dynamics of wild-type Gβ1 and of 1.5 at two sample concentrations (1.1 and 3.0 mM) were investigated by measuring backbone ¹⁵N spin relaxation rate constants. Standard backbone ¹⁵N R₁, R₂ and ¹H-¹⁵N NOE experiments were recorded at 25°C to measure fast (ps–ns) timescale dynamics. In addition, longitudinal (_z) and transverse (_xy) cross-correlation relaxation rate constants were measured to separate contributions to R₂ data from intermediate (µs–ms) timescale conformational exchange (R_ex) and purely fast timescale (R₂°) motions (Kroenke et al., 1998). Refer to the Materials and Methods section for more information on the extraction of R_ex values. Data from residues F3 and I6 in 1.5 were excluded due to spectral overlap, and weak (broadened) resonances prevented measurement of _xy and _z values for residues E27 and T53 in the variant.

Conformational exchange

Figure 4 shows the differences in intermediate timescale (R_ex) protein dynamics between wild-type Gβ1 and the 1.5 variant at both concentrations. Noting that R_ex values >1 s^–1 (above the dotted line in Fig. 4A and C) are considered significant (Hass and Led, 2006), it is clear that wild-type Gβ1 exhibits no notable intermediate timescale backbone motion (Fig. 4A), while both 1.5(H) and 1.5(L) exhibit significant conformational exchange at many points along the entire sequence (Fig. 4C). The difference is illustrated in Fig. 4B, D and E, in which the R_ex values of wild-type, 1.5(H) and 1.5(L) are mapped onto the tertiary structures of each respective protein. These differences are further highlighted in Table I, which lists the average R_ex values over all backbone amides for the wild-type and two 1.5 samples.

View larger version (42K): [in this window] [in a new window] [Download PowerPoint slide]   Fig. 4. Conformational exchange (Rex) in wild-type Gβ1 and in 1.5 at two concentrations. Rex as a function of residue position for (A) wild-type (filled inverted red triangle) and (C) 1.5(H) (filled dark blue triangle) and 1.5(L) (filled light blue circle). The positions of secondary structural elements in the sequence are shown above (A). The dotted line denotes Rex = 1 s–1. Rex values mapped onto the tertiary structure for (B) wild-type, (D) 1.5(H) and (E) 1.5(L), where the Rex value at each position (and, therefore, the magnitude of intermediate timescale motion) is proportional to the depth of color and tube radius; a reference color range is also shown (where a light color is seen at positions with Rex = 0 s–1, and the darkest color corresponds to positions with Rex 7 s–1). Backbone and side chain positions of the six core mutations in 1.5 are indicated in (D) and (E) in orange. (F) Bar graph representing the ratio of Rex values for 1.5(L):1.5(H) as a function of residue position. Data are shown only for residue positions whose Rex values are >1 s–1 [for either or both 1.5(H) or 1.5(L)]. Lightly colored bars represent below average ratio values (i.e. larger than average concentration dependence). The average Rex 1.5(L):1.5(H) value is indicated by the green dotted line. (G) Positions with below average Rex-ratio values are mapped in green onto the Gβ1 tertiary structure. The figure was generated using SigmaPlot (SPSS, 2001) and Pymol (DeLano, 2002).

 

View this table: [in this window] [in a new window]   Table I. Averages and standard deviations for Rex values and spectral density terms

 
One issue to consider is whether R_ex arises from intramolecular conformational exchange, or whether it is merely reflecting the monomer–dimer exchange process. A simple method to assess this is to measure the concentration dependence of R_ex at each position in the sequence (Pfuhl et al., 1999). A bar graph indicating the ratio of the R_ex values for 1.5(L):1.5(H) is shown in Fig. 4F as a function of residue number. Bars that are colored in light green (and segments colored green on the ribbon diagram in Fig. 4G) indicate sequence positions with significant concentration dependence, resulting in lower than average R_ex ratio values; thus, the R_ex values from these positions contain a monomer–dimer exchange component. These residues include 11, 21 and 22 in tight turns, 13, 14 and 16 in the β₂ strand, residues 28, 30, 32 and 36 in the helix and 45 in the β₃ strand. Interestingly, there is also a small subset of residues (numbers 4, 26, 39, 50 and 51) that have increased values of R_ex with decreasing concentration.

If the following relation is applied:

6

where P_tot is the total protein concentration and f_m is the fraction monomer in solution then the monomer:dimer ratio for 1.5(H) is 43:57 and for 1.5(L) is 60:40. The ratios, therefore, indicate that relaxation data recorded at both sample concentrations reflect motion from both the monomer and dimer species of 1.5. Despite this, it is clear that significant differences in protein dynamics are observed between the higher and lower concentration samples, and that these motions primarily reflect dynamics from the more populated species in each sample. For example, it can be seen from the relaxation data (see Supplementary data available at PEDS online, and the results of reduced spectral density mapping in Fig. 5) that the trends in backbone motion for 1.5(L) are a hybrid of the data from wild-type and 1.5(H), with more similarity to the former. Although a lower sample concentration for 1.5(L) would be ideal for more accurate comparison between monomeric 1.5 and wild-type Gβ1 many of the relaxation experiments (especially the cross-correlation relaxation experiments) have low sensitivity and, therefore, could not be measured on samples of 1.5 that were lower than 1 mM.

View larger version (45K): [in this window] [in a new window] [Download PowerPoint slide]   Fig. 5. Reduced spectral density terms: (A) J(0), (B) J(0.87H) and (C) J(N) for wild-type Gβ1 (filled inverted red triangle), 1.5(H) (filled dark blue triangle) and 1.5(L) (filled light blue circle) as a function of residue number. The positions of secondary structural elements are indicated above A. Errors displayed are based on Monte Carlo simulations. Fast timescale backbone motions are mapped onto the tertiary structures of (D) wild-type Gβ1, (E) 1.5(H) and (F) 1.5(L), where the radius and color of the backbone tube are inversely proportional to J(N) at each position (in the narrow range of 0.4–0.6 ns, in order to accentuate the subtle differences). Backbone and side chain positions of the six core mutations in 1.5 are indicated in orange in E and F, and the corresponding positions are indicated for the wild-type in green in D. The figure was generated using SigmaPlot (SPSS, 2001) and Pymol (DeLano, 2002).

 
Reduced spectral density mapping

Data from fast timescale relaxation experiments were analyzed using reduced spectral density mapping (Farrow et al., 1995) and Monte Carlo simulations were used to estimate errors (Spyracopoulos, 2006). Reduced spectral density mapping (Farrow et al., 1995) was applied rather than the more common model-free formalism (Lipari and Szabo, 1982a; b), because the latter must make assumptions about the molecular anisotropy and internal motions of the protein. Since 1.5 is engaged in monomer–dimer exchange in solution, and since the dimer formation may be non-specific, we cannot draw any concrete conclusions on the shape (and, therefore, degree of anisotropic tumbling) or internal motions of 1.5. Thus, reduced spectral density analysis is a more prudent choice, as model-free analysis would be difficult to interpret due to the concentration dependence of the relaxation constant (and, therefore, the order parameter) values (Lefèvre et al., 1996).

J(0), J(0.87_H) and J(_N) data are shown in Fig. 5A–C, respectively, for wild-type Gβ1 as well as 1.5(H) and 1.5(L). Fast timescale protein dynamics are generally indicated by increased values of J(_H) and decreased values of J(0) and J(_N). Significantly increased values of J(0) are normally indicators of intermediate timescale motion; however, our calculations utilized R₂° data as input and, therefore, the J(0) values reported in Fig. 5A (and Supplementary data available at PEDS online) reflect only fast timescale motions. Table I lists average and standard deviation values for all samples. The differences in these averages can be attributed primarily to the oligomerization state of the different proteins: J(0) values increase, J(0.87_H) values decrease and J(_N) values decrease slightly with higher average molecular mass. In general, the trends of J(0) and J(_N) values for the 1.5(L) sample closely match those of the wild-type, while the J(0) values for 1.5(H) are more varied. Values of J(0.87_H) have the same general pattern for all samples, although they fluctuate over a wider range for wild-type Gβ1. The J(_N) values are mapped onto tertiary structures of wild-type (Fig. 5D), 1.5(H) (Fig. 5E) and 1.5(L) (Fig. 5F) to provide structural context for the similarities and differences in fast timescale motions between the protein samples.

As can be seen from Fig. 5D and F, the fast timescale dynamics are similar between wild-type Gβ1 and 1.5(L): both exhibit increased motions near the N-terminus of the β₂ strand (residues 12–14), in the long loop between the helix and β₃ strand (residues 40–41) and in the turn between the β₃ and β₄ strands (residue 48). However, wild-type protein Gβ1 shows spikes of flexibility throughout the β₂ strand at positions 12, 15, 17 and 19; this is especially apparent from the J(0.87_H) data and corresponds to residues for which the backbone amide is not participating in strand-to-strand hydrogen bonding. 1.5 (at both concentrations) also differs from wild-type in the long loop between the helix and β₃ strand since fast timescale dynamics are observed over a larger range of residues (37–41 compared to 40–41). Within the helix, there is little evidence for fast timescale motion although the positions of small observed increases are distributed differently for each sample. However, in both 1.5 samples, values of J(0) and J(_N) near residues 33–35 are much higher than average (Table I), and lower than average in J(0.87_H), indicating that this region of the helix is particularly rigid in the variant at fast timescales. In comparing data from 1.5(L) to that of 1.5(H), both exhibit similar patterns of fast timescale motions throughout the sequence with the exception of minor differences in the loop between the β₂ strand and helix and at a couple of positions along the helix (Fig. 5E and F).

	Discussion

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Support for a 1.5 homodimer rather than a higher-order oligomer

In Fig. 2, the plot of diffusion coefficient values as a function of concentration can be fit to a sigmoidal curve, providing evidence for self-association of 1.5 at NMR concentrations. This is supported by the fact that the maximum and minimum measured diffusion coefficients correlate with known monomeric and dimeric samples of protein Gβ1, respectively: the datum point of D = (3.04 ± 0.12)x10^–6 cm²s^–1 for 0.95 mM 1.5 is consistent with the diffusion coefficient measured for the (monomeric) wild-type protein (D = (2.98 ± 0.13) x 10^–6 cm²s^–1, data not shown), while D = (1.63 ± 0.08) x 10^–6 cm²s^–1 for 2.9 mM 1.5 is similar to the measured diffusion coefficient for a fully dimerised variant of protein Gβ1 (D = (1.72 ± 0.13) x 10^–6 cm²s^–1 for 1.8 mM Gβ1-c3b4, manuscript in preparation). Based on these data we conclude that 1.5 forms a dimer in solution rather than a higher oligomeric state in this range of concentrations.

Conformational exchange in the variant is lacking in wild-type Gβ1

Both 1.5(H) and 1.5(L) exhibit significant intermediate conformational exchange at backbone amide positions throughout the molecule, compared to no notable exchange (no R_ex values >1 s^–1) in the wild-type (Fig. 4A and B). This is supported by a reasonable inverse correlation between R_ex values and peak intensities in the HSQC spectra: 64% of amides (at both 1.5 sample concentrations) with high R_ex (>1 s^–1) have below average peak intensities, while 75% of amides with low or no R_ex (0>R_ex>1 s^–1) have above average peak intensities (data not shown). Additionally, there are no dynamics data for the side chain N1 of W52 because there is no peak for this in the HSQC; this is supportive of intermediate timescale exchange taking place in the core of 1.5 which results in a fully broadened spectral resonance. There is also an inverse correlation between R_ex values and backbone amide protection factors: 69% of the amides with the lowest protection from water (measured using CLEANEX-PM) correspond to amides with high R_ex values (>1 s^–1) (Figs 3B and 4C, respectively). The hydrogen exchange and CLEANEX experiments were originally recorded in an attempt to map the dimer interface in 1.5 (to look for increases in amide protection compared to wild-type), but it is now clear that these experiments reflect the conformational exchange taking place along the backbone of the variant.

Concentration dependent R_ex and CLEANEX data suggest non-specific dimer formation for 1.5

Interpretation of experiments measuring conformational exchange of the 1.5 variant must take into account that each R_ex value may contain a component reflecting monomer–dimer exchange instead of representing only intermediate timescale intramolecular motion. While it is possible that most or all of the R_ex values include a small portion of this exchange, there is a distinct group of residues whose R_ex values have considerable concentration dependence (Fig. 4F). However, rather than seeing these residues map to a single specific region of the tertiary structure we instead observe that one subset of residues (11, 13, 14 and 36) clusters in one general region to form a loose dimer interface, while other residues (21, 22, 28, 30–32 and 45) are scattered throughout the molecule (Fig. 4G).

In comparing the results from CLEANEX experiments recorded on a 0.98 mM sample (data not shown) to the 2.6 mM sample of 1.5 (Fig. 3B), there are several residues for which the amide protection factor is concentration dependent, suggesting that the residues are in or near a dimer interface: protection factors decrease with decreasing concentration at positions 9, 10, 16 and 35–37. All of these residues (except 16) map to the region of the molecule that includes the C-terminus of the helix and the loop between strands β₁ and β₂. Analysis of concentration dependence from CLEANEX data is more informative in this case than data from standard H/D exchange experiments because the former reflect fast-exchanging protons on the timescale of seconds and would be much more likely to capture the formation and dissociation of transient interface interactions than H/D exchange, which reflects slow self-association events on the timescale of minutes.

The region of 1.5 in which residues show concentration dependence in CLEANEX is also observed in R_ex experiments. However, as has already been mentioned, there are additional concentration dependent residues seen only in the R_ex data which are scattered in several regions throughout the molecule. It is possible that the R_ex data highlight multiple sites of non-specific self-association for the 1.5 variant that are populated very briefly (the most transient of which are probed only in the µs–ms timescale regime), while the CLEANEX data represent the most common interface site which is populated for the longest period of time and, therefore, can be seen in the timescale of the CLEANEX experiments in addition to the faster timescale of the R_ex experiments. Kinetics experiments might be helpful in elucidating the k_on and k_off of the interaction to confirm this hypothesis.

One possible reason for the dimer interface (near the β₁/β₂ loop and helix C-terminus) that can be observed in both CLEANEX and R_ex data is that the hydrophobic surface of residue L12 is 30% less buried in 1.5 compared to wild-type, as calculated by the program QSURF, which is a component of the ORBIT software package. This local increase in hydrophobicity on the surface of the molecule could be enough to promote weak self-association in this region.

Although we have determined that R_ex in the β₂ strand region is mainly reflective of monomer–dimer exchange, the extremely low protection factors from H/D exchange experiments indicate that this region is also involved in some slow intramolecular exchange. Using the equation G°_HX=RT ln(PF) (where PF is the protection factor for a given amide group) one can calculate an average G for local and global conformational equilibria (that allow hydrogen exchange to take place) (Lee et al., 2005). The average for wild-type protein Gβ1 over the entire molecule can be estimated at G°_HX22 kJ/mol, while for 1.5 G°_HX 14 kJ/mol, confirming there is a significantly lower barrier to conformational exchange for the variant compared to wild-type.

For a few residues (4, 26, 39, 50 and 51), R_ex increases with decreased concentration. Previous researchers have suggested that positions within a protein that undergo reduction of intramolecular conformational exchange upon dimerization are close to the dimer interface (Åkerud et al., 2002). However, in analyzing the data from 1.5, we see that these residues are directly adjacent to positions with some of the lowest levels of concentration dependence in R_ex and some of the highest values of R_ex. Since Gβ1 is a small domain, residues with this inverse concentration dependence may be reflecting the transmitted effects of dimer formation and dissociation rather than being directly involved in the dimer interface. This scenario is supported by experiments by Mayer et al. (Mayer et al., 2003) that have illustrated the existence of a network of short- and long-range correlated motions throughout the Gβ1 backbone.

Conformational exchange in 1.5 is localized in two regions

The (non-concentration-dependent) data from experiments measuring intermediate timescale (µs–ms) protein dynamics have highlighted an intriguing correlation between core overpacking and conformational exchange. Backbone amide groups in 1.5 with the highest R_ex values cluster primarily in two areas: the first area encompasses the long loop between the helix and β₃ strand, and surrounds the V39I and A34I mutations in the variant (located on the loop and at the C-terminal end of the helix, respectively) (Fig. 4D and E). In the latter mutation, the side chain volume is almost doubled compared to wild-type. In the NMR structure the Ile34 side chain comes into close contact with the side chains of residues 40 and 54 and this increased local crowding may, therefore, be a major cause of the observed intermediate timescale backbone motions for these residues. Future studies to characterize the protein dynamics of a variant without the A34I mutation may be able to confirm this assertion. It is important to note that, although the close packing is experienced by the side chains, it is easy to see how the effects of side chain overpacking might be transmitted to the backbone and therefore result in the observed backbone motions. Future experiments that monitor motions of side chain methyl groups in this region will be used to support the observed backbone dynamics.

Residues with increased backbone conformational exchange in the second region (near the N-terminus of the helix) are found at positions surrounding the F52W mutation (including 4, 22, 24, 25 and 51, where the Trp side chain is in close proximity to the backbone at these positions and 43, where the two Trp side chains are in contact). Additionally, residues E27 (located on the helix directly across from W52) and T53 have such broadened peak intensity that we cannot report on hydrogen exchange data for residue 27 and on R_ex data for both 27 and 53, indicative of significant intermediate timescale conformational exchange in that region. Thus, while a larger side chain such as W52 can be accommodated in the core of a Gβ1-fold, it may be causing increased strain that is only relieved by the observed intermediate timescale dynamics. An additional reason for the increase in conformational exchange in this region may be due to the loss of an apparent aromatic T-stacking interaction between F30 and F52, which is replaced by mutations F30L and F52W in the 1.5 variant. This stacking interaction may have been an important source of rigid support within the core of wild-type Gβ1 that is eliminated in the variant (Burley and Petsko, 1985). A third (previously observed) potential source for increased core motions is that the design of 1.5 did not provide for a hydrogen-bonding partner for the imino group on the Trp side chain at position 52 (Ross et al., 2001).

In general, despite overpacking at various positions within the core of the 1.5 variant, the protein is still able to fold and is relatively thermostable. It is likely that the protein is able to remain folded with an overpacked core only by interchanging between various conformations, each of which offsets the overpacking in a slightly different way. It is this ensemble of conformations that is being observed indirectly through the R_ex and, to a lesser extent, hydrogen exchange data.

A general decrease in fast timescale dynamics in 1.5

Although there are significant differences in intermediate (µs–ms) timescale backbone amide dynamics between wild-type Gβ1 and the 1.5 variant, there are only small differences between the two proteins when assessing fast timescale (ps–ns) motions. Overall, the region surrounding the F52W mutation in 1.5 seems to be more rigid at fast timescales, while the region around the A34I mutation shows evidence for some increased motion compared to wild-type. It is possible that, when probed for fast timescale motion, the increased rigidity in residues close to F52W may reflect solely side chain overpacking, while the impact of other factors (such as the loss of an aromatic T-stacking interaction) may only be detected at the slower timescale. Indeed, the loss in 1.5 of the alternating trends in backbone dynamics in the β₂ strand (Fig. 5B) might also reflect overpacking in the region. As in the wild-type Gβ1, there are backbone amides in 1.5 that are not participating in cross-strand hydrogen bonding; however, observation of increased motions at these positions compared to hydrogen bonded positions may be dampened in 1.5 by the overpacked side chains. The increased fast timescale motion for 1.5 near A34I could be explained by the lower degree of defined secondary structure in that region: many of the side chains affected by this mutation are in the long loop between the helix and β₃ strand, so a wider range of timescales of motion may be available. In contrast, residues in the region near F52W are primarily found in secondary structural elements, which could restrict motion at fast timescales but still permit more global conformational exchange at slower timescales.

Assessing the relationship between the dynamics data and the variant structure

The original objective for the project that resulted in the 1.5 variant was to design a sequence with a tertiary structure whose helix was translated by 1.5 Å farther away from the β₂-sheet than its position in wild-type protein Gβ1. When the solution structure of the variant was solved, however, the fold had a backbone RMSD to the wild-type of 1.03 Å, compared with a RMSD of 1.26 Å to the design target backbone (Ross et al., 2001). In looking at the core mutations that were obtained through the computational design process, the A34I and F52W mutations may have been the two most crucial changes needed to achieve the desired helix translation, since the other four mutations in 1.5 are conservative and do not appreciably increase the bulk in the core of the protein. As discussed above, the mutations at 34 and 52 introduce significant increases in side chain volume. It is, therefore, likely that these two mutations were expected to push the helix away from the β-sheet, since each is strategically placed in range of one end of the helix (A34I at the C-terminus and F52W pointing toward the N-terminus). Considering that the desired translation distance was relatively small, it would seem feasible that two carefully placed mutations would work. However, the design did not permit for an extension of the linker length at each end of the helix, particularly in the tight turn between β₂ and the N-terminus. This resulted in a restriction of the distance that the helix could be translated away from the β-sheet and suggests why the resulting NMR solution structure shows little displacement (Ross et al., 2001) and the two side chains have contributed to the core overpacking. Not surprisingly, it appears that the A34I and F52W mutations are located in the two regions of the molecule that show the majority of the changes in fast timescale backbone motions, intermediate timescale conformational exchange and decreases in amide proton protection factors. It may be considered unusual that conformational exchange has increased so drastically with these increases in side chain volume; in most previous studies, researchers have decreased side chain volume in order to probe increased conformational exchange (Hanson et al., 2003; Jarymowycz and Stone, 2006). However, it is likely the combination of core overpacking, non-specific hydrophobic–hydrophobic core contacts and the loss of the aromatic ring T-stacking interaction in the variant can explain why we observe the opposite trend.

The correlation between stability, packing and conformational flexibility

How do we reconcile an increase in the volume of buried hydrophobic groups and an increase in conformational flexibility (and presumably increased entropy in the folded protein) with a decrease in thermal stability in the 1.5 variant (T_m = 73°C) compared to wild-type (T_m = 85°C) (Su and Mayo, 1997), when others have observed a positive correlation between increased flexibility and increased thermal stability (Wand, 2001)? Several years ago Ventura et al. (Ventura et al., 2002) explored the cost of ‘conformational strain’ in the core of a redesigned spectrin SH3 variant. Although tight packing of a hydrophobic core and increased burial of hydrophobic groups in the folded protein generally creates a more stable protein, it was found that the stability gain could be offset if the design resulted in extra side chain bulk that overpacked the core. The NMR structure of 1.5 does appear to have an overpacked core, as evidenced by the fact that the design has 1.11 the fraction of core side chain volume of wild-type (Su and Mayo, 1997), but has a similar tertiary fold. It is likely that these factors have been sufficient to tip the fine balance between the benefits of tight packing and the cost of overpacking and give a protein with a lower (albeit still quite thermostable) melting temperature. The decreased thermostability of 1.5 compared to wild-type can be further explained by work carried out by Tang and Dill (Tang and Dill, 1998), who found that protein stability is inversely correlated with flexibility that arises from larger fluctuations (such as those reflected in R_ex data or the decreased protection factors for the variant).

Implications for protein design

Although the design of core variant 1.5 is the product of a test exercise, the results of the analyses of its structure and backbone protein dynamics highlight the possible consequences of incorporating major changes to the identities of amino acids within a given sequence. At this point the data indicate that core overpacking and/or increased non-specific core interactions are the major causes of the increased conformational exchange in the 1.5 variant of protein Gβ1. Considering how important µs–ms timescale motion is in biological function (particularly in enzyme activity) (Eisenmesser et al., 2005), it would, therefore, be worth exploring both of these as possible design strategies. Support for the concept that non-specific interactions may cause intermediate timescale motions has recently arisen in a paper by Foulkes-Murzycki et al. (Foulkes-Murzycki et al., 2007), who discuss ‘hydrophobic sliding’ in the core of HIV-1 protease and postulate that this could be a mechanism for protein dynamics. If this is the case, purposeful introduction of non-specific core interactions could be a method for introducing µs–ms timescale motions into protein designs, although care would need to be taken to permit increased dynamics while maintaining a well-folded and stable protein.

In the case of the 1.5 variant, it is interesting that the outcome of an effort to achieve one computational design goal has caused significant unintended side effects within the molecule. While the positive or negative implications of these effects are still to be determined, we must remain fully aware that these side effects are possible (and perhaps likely). These studies have helped to emphasize the need to characterize in more detail the biophysical properties of designed proteins in order to better understand and predict how mutations will affect the molecule, with future applications in the design of proteins that carry out biological functions.

	Supplementary data

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Supplementary data, including R₁, R₂° (transverse relaxation rate constant that is free of chemical/conformational exchange contributions), NOE, _xy, _z and R_ex data are provided for wild-type protein Gβ1 and for the 1.5 variant at two concentrations, and are available at PEDS online.

	Funding

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Natural Sciences and Engineering Research Council of Canada (Postdoctoral Fellowship to K.A.C.); Howard Hughes Medical Institute.

	Footnotes

 
³ Present address: Department of Chemistry and Biochemistry, California State University Northridge, Northridge, CA 91330-8262, USA.

Abbreviations and symbols: CLEANEX-PM, phase-modulated CLEAN chemical exchange; DSS, 2,2-dimethyl-2-silapentane-5-sulfonate sodium salt; 1.5, a 6-fold variant of core residues of Gβ1; 1.5(H), a higher concentration sample of the variant at 3.0 mM; 1.5(L), a lower concentration sample of the variant at 1.1 mM; HSQC, heteronuclear single-quantum coherence spectroscopy; IPTG, isopropyl β-D-1-thiogalactopyranoside; J(0), J(0.87_H) and J(_N), spectral density functions derived at zero, proton and nitrogen frequencies, respectively; NOE, nuclear Overhauser enhancement; R₁, longitudinal relaxation rate constant; R₂, transverse relaxation rate constant; R₂°, exchange-free transverse relaxation rate constant; R_ex, chemical (conformational) exchange contribution to R₂.

Edited by Frances H. Arnold

	Acknowledgements

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
The authors would like to thank Drs Julie Forman-Kay, John Love and Kevin Plaxco for insightful comments on the manuscript, as well as Drs Andrew Chong, Kevin Gardner, Lewis Kay, Dorothee Kern, Gregory Lee and Leo Spyracopoulos for advice on experiments and data analysis. We are also grateful to Dr Scott Ross for NMR assistance and Mr. Ben Allen for his computational expertise.

	References

Top Abstract Introduction Materials and methods Results Discussion Supplementary data Funding Acknowledgements References

 
Åkerud T., Thulin E., Van Etten R.L., Akke M. J. Mol. Biol. (2002) 322:137–152.[CrossRef][Web of Science][Medline]

Bai Y., Milne J.S., Mayne L., Englander S.W. Proteins (1993) 17:75–86.[CrossRef][Web of Science][Medline]

Burley S.K., Petsko G.A. Science (1985) 229:23–28.[Abstract/Free Full Text]

Choy W.Y., Mulder F.A., Crowhurst K.A., Muhandiram D.R., Millett I.S., Doniach S., Forman-Kay J.D., Kay L.E. J. Mol. Biol. (2002) 316:101–112.[CrossRef][Web of Science][Medline]

Dahiyat B.I., Mayo S.L. Science (1997) 278:82–87.[Abstract/Free Full Text]

Delaglio F. (1995) NIH Laboratory of Chemical Physics: Bethesda, MD. NMRDraw (v. 1.8) [Software].

DeLano W.L. (2002) The PyMOL Molecular Graphics System (v. 0.99) [Software]. DeLano Scientific: Palo Alto, CA.

Dempsey C.E. Prog. Nucl. Mag. Res. Sp. (2001) 39:135–170.[CrossRef]

Desjarlais J.R., Handel T.M. J. Mol. Biol. (1999) 290:305–318.[CrossRef][Web of Science][Medline]

Dobson N., Dantas G., Baker D., Varani G. Structure (2006) 14:847–856.[Medline]

Eisenmesser E.Z., Millet O., Labeikovsky W., Korzhnev D.M., Wolf-Watz M., Bosco D.A., Skalicky J.J., Kay L.E., Kern D. Nature (2005) 438:117–121.[CrossRef][Web of Science][Medline]

Farrow N.A., Muhandiram R., Singer A.U., Pascal S.M., Kay C.M., Gish G., Shoelson S.E., Pawson T., Forman-Kay J.D., Kay L.E. Biochemistry (1994) 33:5984–6003.[CrossRef][Web of Science][Medline]

Farrow N.A., Zhang O., Szabo A., Torchia D.A., Kay L.E. J. Biomol. NMR (1995) 6:153–162.[Web of Science][Medline]

Foulkes-Murzycki J.E., Scott W.R., Schiffer C.A. Structure (2007) 15:225–233.[Medline]

Gillespie B., Vu D.M., Shah P.S., Marshall S.A., Dyer R.B., Mayo S.L., Plaxco K.W. J. Mol. Biol. (2003) 330:813–819.[CrossRef][Web of Science][Medline]

Hanson W.M., Beeser S.A., Oas T.G., Goldenberg D.P. J. Mol. Biol. (2003) 333:425–441.[CrossRef][Web of Science][Medline]

Harbury P.B., Plecs J.J., Tidor B., Alber T., Kim P.S. Science (1998) 282:1462–1467.[Abstract/Free Full Text]

Hass M.A., Led J.J. Magn. Reson. Chem. (2006) 44:761–769.[CrossRef][Web of Science][Medline]

Hwang T.-L., Mori S., Shaka A.J. J. Am. Chem. Soc. (1997) 119:6203–6204.[CrossRef][Web of Science]

Hwang T.-L., van Zijl P.C.M., Mori S. J. Biomol. NMR (1998) 11:221–226.[CrossRef][Web of Science][Medline]

Jarymowycz V.A., Stone M.J. Chem. Rev. (2006) 106:1624–1671.[CrossRef][Web of Science][Medline]

Johnson B.A., Blevins R.A. J. Biomol. NMR (1994) a 4:603–614.[CrossRef][Web of Science]

Johnson B.A., Blevins R. (1997) Whitehouse Station, NJ: Merck and Co. 4.1.2 ed.;.

Johnson B.H., Hecht M.H. Bio/Technology (1994) b 12:1357–1360.[CrossRef][Medline]

Johnson E.C., Lazar G.A., Desjarlais J.R., Handel T.M. Structure (1999) 7:967–976.[Medline]

Kay L.E., Keifer P., Saarinen T. J. Am. Chem. Soc. (1992) 114:10663–10665.[CrossRef][Web of Science]

Kroenke C.D., Loria J.P., Lee L.K., Rance M., Palmer A.G. III. J. Am. Chem. Soc. (1998) 120:7905–7915.[CrossRef][Web of Science]

Kuhlman B., Dantas G., Ireton G.C., Varani G., Stoddard B.L., Baker D. Science (2003) 302:1364–1368.[Abstract/Free Full Text]

Larson S.M., England J.L., Desjarlais J.R., Pande V.S. Protein Sci. (2002) 11:2804–2813.[CrossRef][Web of Science][Medline]

Lee G.M., Donaldson L.W., Pufall M.A., Kang H.S., Pot I., Graves B.J., McIntosh L.P. J. Biol. Chem. (2005) 280:7088–7099.[Abstract/Free Full Text]

Lefèvre J.-F., Dayie K.T., Peng J.W., Wagner G. Biochemistry (1996) 35:2674–2686.[CrossRef][Web of Science][Medline]

Lipari G., Szabo A. J. Am. Chem. Soc. (1982) a 104:4546–4559.[CrossRef][Web of Science]

Lipari G., Szabo A. J. Am. Chem. Soc. (1982) b 104:4559–4570.[CrossRef][Web of Science]

Looger L.L., Dwyer M.A., Smith J.J., Hellinga H.W. Nature (2003) 423:185–190.[CrossRef][Web of Science][Medline]

Mayer K.L., Earley M.R., Gupta S., Pichumani K., Regan L., Stone M.J. Nat. Struct. Biol. (2003) 10:962–965.[CrossRef][Web of Science][Medline]

McCammon J.A. Biochim. Biophys. Acta (2005) 1754:221–224.[Medline]

Muhandiram D.R., Kay L.E. J. Magn. Reson. Ser. B (1994) 103:203–216.[CrossRef][Web of Science]

Palmer A.G. III. (1998) New York: Columbia University. 1.3 ed.;.

Pfuhl M., Chen H.A., Kristensen S.M., Driscoll P.C. J. Biomol. NMR (1999) 14:307–320.[CrossRef][Web of Science][Medline]

Ross S.A., Sarisky C.A., Su A., Mayo S.L. Protein Sci. (2001) 10:450–454.[CrossRef][Web of Science][Medline]

Santoro J., King G.C. J. Magn. Reson. (1992) 97:202–207.[Web of Science]

SPSS. (2001) Systat Software, Inc: San Jose, CA. SigmaPlot (v. 7.0) [Software].

Spyracopoulos L. J. Biomol. NMR (2006) 36:215–224.[CrossRef][Web of Science][Medline]

Su A., Mayo S.L. Protein Sci. (1997) 6:1701–1707.[Web of Science][Medline]

Tang K.E., Dill K.A. J. Biomol. Struct. Dyn. (1998) 16:397–411.[Web of Science][Medline]

Ventura S., Vega M.C., Lacroix E., Angrand I., Spagnolo L., Serrano L. Nat. Struct. Biol. (2002) 9:485–493.[CrossRef][Web of Science][Medline]

Walsh S.T.R., Lee A.L., DeGrado W.F., Wand A.J. Biochemistry (2001) 40:9560–9569.[CrossRef][Web of Science][Medline]

Wand A.J. Nat. Struct. Biol. (2001) 8:926–931.[CrossRef][Web of Science][Medline]

Wolfram Research. (2005) Wolfram Research, Inc: Champaign, Illinois. Mathematica (v. 5.2) [Software].

Zavodszky M.I., Lei M., Thorpe M.F., Day A.R., Kuhn L.A. Proteins (2004) 57:243–261.[CrossRef][Web of Science][Medline]

Zwahlen C., Legault P., Vincent S.J.F., Greenblatt J., Konrat R., Kay L.E. J. Am. Chem. Soc. (1997) 119:6711–6721.[CrossRef][Web of Science]

Received May 21, 2008; revised May 21, 2008; accepted May 23, 2008.

CiteULike    Connotea    Del.icio.us    What's this?

This Article Abstract FREE Full Text (PDF) Supplementary Data All Versions of this Article: 21/9/577    most recent gzn035v1 Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Crowhurst, K. A. Articles by Mayo, S. L. Search for Related Content PubMed PubMed Citation Articles by Crowhurst, K. A. Articles by Mayo, S. L. Social Bookmarking          What's this?

Online ISSN 1741-0134 - Print ISSN 1741-0126

Copyright © 2009 Oxford University Press

Oxford Journals Oxford University Press

• Site Map
• Privacy Policy
• Frequently Asked Questions

Other Oxford University Press sites: Oxford University Press Oxford Journals China Oxford Journals Japan American National Biography Booksellers' Information Service Children's Fiction and Poetry Children's Reference Corporate & Special Sales Dictionaries Dictionary of National Biography Digital Reference English Language Teaching Higher Education Textbooks Humanities International Education Unit Law Medicine Music Online Products Oxford English Dictionary Reference Rights and Permissions Science School Books Social Sciences Very Short Introductions World's Classics