Rotameric preferences of a protein spin label at edge‐strand β‐sheet sites

Abstract

Protein spin labeling to yield the nitroxide‐based R1 side chain is a powerful method to measure protein dynamics and structure by electron spin resonance. However, R1 measurements are complicated by the flexibility of the side chain. While analysis approaches for solvent‐exposed α‐helical environment have been developed to partially account for flexibility, similar work in β‐sheets is lacking. The goal of this study is to provide the first essential steps for understanding the conformational preferences of R1 within edge β‐strands using X‐ray crystallography and double electron electron resonance (DEER) distance measurements. Crystal structures yielded seven rotamers for a non‐hydrogen‐bonded site and three rotamers for a hydrogen‐bonded site. The observed rotamers indicate contextual differences in R1 conformational preferences compared to other solvent‐exposed environments. For the DEER measurements, each strand site was paired with the same α‐helical site elsewhere on the protein. The most probable distance observed by DEER is rationalized based on the rotamers observed in the crystal structure. Additionally, the appropriateness of common molecular modeling methods that account for R1 conformational preferences are assessed for the β‐sheet environment. These results show that interpretation of R1 behavior in β‐sheets is difficult and indicate further development is needed for these computational methods to correctly relate DEER distances to protein structure at edge β‐strand sites.

Short abstract

PDB Code(s): 5BMG; 5BMH; 5BMI

Abbreviations

SDSL

site‐directed spin labeling

ESR

electron spin resonance spectroscopy

NMR

nuclear magnetic resonance spectroscopy

MTSSL

methanothiosulfonate spin label

R1

resulting side chain from the reaction of MTSSL with free cysteine

DEER

double electron electron resonance

GB1

B1 immunoglobulin‐binding domain of protein G

WT

wild type

MD

molecular dynamics.

Introduction

Incorporation of unpaired electron spins into proteins is an increasingly useful tool for elucidation of protein structure and dynamics using magnetic resonance techniques.1, 2, 3 This process, known as site‐directed spin labeling (SDSL), involves the site‐specific placement of either stable organic radicals or paramagnetic metals within a protein. While metals such as Mn²⁺, Cu²⁺, and Gd³⁺ show great promise in recent work,4, 5, 6, 7, 8, 9, 10 the most commonly utilized spin probe is the nitroxide‐functionalized side‐chain R1. R1 is generated by reacting the thiol‐specific methanothiosulfonate spin label (MTSSL) with cysteine residues. The resulting R1 residue [Fig. 1(A)] is a useful structural probe that has been shown to provide site‐specific dynamics information in some protein environments11 as well as long‐range distance constraints for proteins and protein complexes.1, 12

Figure 1.

(A) The common R1 spin label and the five associated dihedral angles that define R1 rotameric states. (B) Cartoon displaying the contextual differences between edge β‐strand sites, where the residue in question is (15R1) or is not (44R1) hydrogen bonded to the neighboring strand. Non‐hydrogen bonded sites are expected to be more prone to clashes with the neighboring chain because of the direction of the C_α–C_β vector.

Nanometer‐scale distances can be routinely measured between multiple R1 sites using electron spin resonance (ESR) spectroscopy, providing valuable long range (15–100 Å) distance constraints.13, 14, 15 An important limitation in such measurements arises from the fact that the measured distance is between the N—O bonds of the nitroxide rings. This gives rise to a fundamental issue complicating the use of R1 in structural biology: the functional probe in the magnetic resonance experiments (the N—O bonds of the nitroxide ring) is separated from the structural feature of interest (the protein backbone) by five freely rotatable single bonds [Fig. 1(A)]. This rapid fluctuation in the dihedral values of each of the bonds (designated χ ₁ through χ ₅) leads to a high degree of flexibility of the side chain, and a range of nitroxide ring locations relative to the attachment site.

Side‐chain motions dominate distance distributions between R1 pairs and make it challenging to extract C_α—C_α distances from such measurements. For example, we previously generated quantitative comparisons of the C_α—C_α distribution within a protein–DNA complex by comparison of distance distributions obtained by double electron electron resonance (DEER)16 and molecular dynamics (MD) simulations; the C_α—C_α distribution from the simulation had a width six times narrower than the DEER experiment.17 In an effort to account for R1 flexibility in distance measurements, various computational tools have been developed.18, 19, 20, 21, 22 While useful in some contexts, these tools cannot consistently account for R1 flexibility in all cases. Consequently, a number of DEER measurements are typically required before quantitative conclusions about assembly structure or protein conformation can be drawn.17, 23, 24, 25, 26, 27 One important limitation of most of the above tools is that they were constructed by considering R1 free in solution19 or R1 in α‐helices.28, 29

Given the prevalence of β‐sheets in protein structure, broadening the scope of available methods for the analysis of R1 conformational behavior to account for this environment is an important goal. A previous X‐band continuous wave (CW) ESR dynamics study of R1 in a variety of solvent‐exposed β‐sheet locations suggested that its rotameric behavior can be dominated by system‐specific local interactions.30 We also incorporated R1 at a β‐sheet site in a protein–DNA complex;16, 17 DEER measurements were able to generate information about the C_α—C_α distance distributions, but full MD simulations were necessary to account for the local factors that governed rotameric preferences. To gain further insight into R1 behavior in β‐sheet environments, we later explored an internal strand site with a variety of techniques and identified R1 rotamers not previously observed.31 These results indicated that methods currently used to interpret site‐specific dynamics information using R1 in α‐helices are not reliable for interior‐strand β‐sheet environments. Collectively, the above data suggest that β‐sheets have complex effects on R1 rotameric preferences and prediction of its behavior in these contexts is challenging.

The conformational behavior of solvent‐exposed side chains at interior strands in β‐sheets is strongly influenced by local interactions, both from nearby residues in the same strand as well as residues at the two flanking strands. In contrast, edge‐strand side chains in a sheet have a higher likelihood of being context‐independent in their rotameric preferences due to the loss of one of the neighboring strands as a potential source of interference. Edge‐strand sites can be divided into two categories based on whether or not the residue in question forms backbone hydrogen bonds to the sole neighboring strand [Fig. 1(B)]. This distinction is important because it also dictates the orientation of the C_α—C_β bond relative to the rest of the β‐sheet. When the backbone C=O and N—H bonds of a residue are not hydrogen bonded to the neighboring strand, the C_α—C_β bond points slightly toward the sheet, increasing the possibility of residues on the neighboring chain affecting side‐chain rotameric preferences. All edge sites previously explored in the X‐band CW study of R1 fall into this non‐hydrogen‐bonding category.30 When the backbone of a residue at an edge‐strand is hydrogen bonded to the neighboring strand, the C_α—C_β bond points slightly away from the rest of the sheet. As such, the side‐chain is less likely to be influenced by local contacts.

The above analysis led us to hypothesize that R1 in hydrogen‐bonding edge‐strand β‐sheet sites may be able to more fully sample side‐chain conformational space and, in turn, find intra‐residue interactions that restrict flexibility, such as observed in α‐helix contexts. Here, we test this hypothesis by systematically comparing R1 in two types of solvent‐exposed edge‐strand environments. We explore differences in conformational preferences at a non‐hydrogen‐bonding site and a hydrogen‐bonding site in a β‐sheet of the B1 immunoglobulin‐binding domain of protein G (GB1)31 through a variety of methods. X‐ray crystal structures show rotamers of R1 that differ between the two sites and are unlike any previously observed R1 conformers.31, 32, 33, 34 DEER distance measurements were also obtained involving each β‐sheet site in two double‐labeled proteins. Molecular modeling of the most probable distance was performed to allow analysis of R1 behavior in solution as compared to the crystal lattice. The appropriateness of common computational tools used for predicting DEER distances in these two edge strand environments was also assessed. Taken together, the results presented here demonstrate that R1 rotameric behavior in edge strands of solvent exposed β‐sheets depend strongly on whether the strand bearing R1 is hydrogen bonded to the neighboring strand.

Results and Discussion

Spin‐labeled protein design

The goal of exploring edge strand environments was accomplished through use of the protein GB1, which contains a single, four‐stranded β‐sheet. Two sites in the GB1 β‐sheet were modified to explore different edge‐strand environments: E15 (backbone not hydrogen‐bonded to the neighboring strand) and T44 (backbone hydrogen‐bonded to the neighboring strand). Both of these positions are located midway along the GB1 β‐sheet which is ∼6 residues wide. E15 is expected to project the C_α—C_β vector of the side chain into the neighboring strand such that sterics may dictate rotameric preferences, while T44 is expected to project the C_α—C_β vector of the side chain away from the protein [Fig. 1(B)].

Cysteine mutations were incorporated at the desired sites (i.e., E15C and T44C) and subsequently labeled by standard methods31 to produce GB1 mutants 15R1 and 44R1 for crystallization experiments. As a means to further characterize R1 at these sites, distance measurements were performed using DEER on GB1 mutants pairing 15R1 or 44R1 with a second label at a solvent exposed site (28R1) on the α‐helix at the opposite face of the protein. We have shown previously that the 28R1 mutant of GB1 displays side‐chain rotameric preferences typical for R1 in solvent‐exposed α‐helices.31 Furthermore, computational tools18, 19, 20, 21, 22 that can account for R1 flexibility have been found useful to interpret distance measurements at helical sites. Thus, the double mutants E15C/K28C and K28C/T44C were generated and in turn labeled to yield 15R1/28R1 and 28R1/44R1.

X‐ray crystallographic analysis

In an effort to obtain direct information about R1 side‐chain rotameric preferences at edge‐strand sites, we grew crystals of 15R1 and 44R1 by hanging‐drop vapor diffusion and collected X‐ray diffractions data for three distinct crystal forms (one of 15R1 and two of 44R1; Table 1). The structure of the 15R1 mutant was refined to 2.2 Å, and eight copies of the protein were observed in the asymmetric unit. Each monomer exhibited a fold very similar to wild‐type GB1. The eight monomers in the asymmetric unit provided eight crystallographically independent R1 sites to examine, and seven of the sites contained side chains fully resolved in electron density (Fig. 2 and Table 2). The individual rotamers observed at these sites are referred to as 15R1₁−15R1₈; the only rotamer where all five dihedral angles are not fully resolved is 15R1₃.

Table 1.

Data Collection and Refinement Statistics for the Three Crystal Structures

Protein	15R1	44R1A	44R1B
Data collection
Unit cell dimensions (Å, °)	a = 52.3, b = 79.5, c = 52.4 α = γ = 90, β = 90.1	a = 25.0, b = 37.5, c = 26.4 α = γ = 90, β = 108	a = 25.0, b = 37.2, c = 49.2 α = β = γ = 90
Space group	P21	P21	P212121
Resolution (Å)	27.11–2.20 (2.28–2.20)	25.03–2.50 (2.59–2.50)	25.05–1.60 (1.66–1.60)
Total observations	66,931	5,475	38,883
Unique observations	21,244	1,575	6,458
Redundancy	3.2 (2.5)	3.5 (3.6)	6.0 (4.4)
Completeness (%)	97.1 (90.9)	96.1 (95.9)	100 (97.2)
I/σ	9.1 (3.7)	13.6 (2.3)	24.0 (3.1)
R merge (%)	8.4 (22.0)	6.4 (22.8)	5.2 (31.6)
Refinement
Resolution (Å)	27.08–2.20	25.03–2.50	24.6–1.60
R (%)	18.9	23.6	15.6
R free (%)	21.8	25.1	17.5
Avg. B factor (Å2)	33.6	49.9	19.3
RMSD
Bonds (Å)	0.007	0.003	0.014
Angles (°)	1.3	1.03	1.4

Figure 2.

R1 side chains from the 15R1 crystal structure. Conformers are categorized and grouped based on the χ ₁/χ ₂ values.

Table 2.

Summary of R1 Rotamers Observed in the Crystal Structures

Rotamer	Conformation	χ 1	χ 2	χ 3	χ 4	χ 5	Cα—Sδ Dist. (Å)
15R11	{t,t?}	−171	155	131	120	179	4.4
15R12	{p,p?}	69	104	133	−79	26	4.0
15R13	{t?,−}	135	–	–	–	–	–
15R14	{p,p?}	66	89	132	−67	47	3.8
15R15	{t,t}	171	−167	−68	−109	−51	4.5
15R16	{p,t?}	81	135	93	175	−63	4.3
15R17	{t?,t?}	−151	−145	135	129	127	4.3
15R18	{t?,t}	144	−177	−90	−157	70	4.5
44R1A1	{t,p?}	−163	101	72	106	28	3.9
44R1B1	{t,t?}	−166	152	69	45	−104	4.4
44R1B2	{m,t?}	−61	−153	−75	96	−100	4.4

The final column provides the distance between the C_α and S_δ for each of the observed R1 rotamers (see text).

Two distinct crystal forms of the 44R1‐GB1 mutant were obtained in crystallization experiments. The structure of one form, designated 44R1_A, was refined to 2.5 Å, and that of the other, designated 44R1_B, was refined to 1.6 Å. As with the 15R1 mutant, the overall folds of 44R1 observed in both structures were virtually identical to wild‐type GB1. The asymmetric unit of 44R1_A contains a single copy of the protein with a single fully resolved R1 side‐chain rotamer, 44R1_A1 (Fig. 3 and Table 2). The 44R1_B crystal form also contains a single GB1 monomer, but the R1 side chain was found to populate two partially occupied rotamers, 44R1_B1 and 44R1_B2 (Fig. 3 and Table 2).

Figure 3.

R1 side chains from the structures of the two different crystal forms of 44R1. Conformers are categorized based on the χ ₁/χ ₂ values.

An important consideration in applying X‐ray crystallography to investigate R1 conformational preferences is determining whether the context of the crystal lattice influences the R1 residue. Contacts with other protein chains in the asymmetric unit or symmetry‐related copies in the crystal have the potential to bias side‐chain torsional angles in the same way as intramolecular contacts with other residues in the protein. Although ten crystallographically independent R1 rotamers were observed in the three structures described above, inspection of the local environments around each site revealed that in all cases R1 was likely influenced by intermolecular contacts involving other protein copies in the crystal lattice. Despite this limitation, the structural data obtained are useful in elucidating fundamental behavior of R1.

A rotamer of R1 is defined by the five dihedral angles, χ ₁–χ ₅, about the five rotatable bonds between the protein backbone and the nitroxide ring [Fig. 1(A)]. Dihedral angles χ ₁ to χ ₃ depend primarily on “native” contacts involving local residues in the same protein chain, whereas χ ₄ and χ ₅ are more prone to influence by “non‐native” contacts involving crystal packing. In crystal structures of R1 at solvent‐exposed α‐helix sites, the side chain is only resolved in electron density to a point that enables determination of χ ₁–χ ₃; this is usually rationalized as resulting from highly dynamic rotation about χ ₄ and χ ₅.32, 34 Nevertheless, χ ₁—χ ₃ contain essential information about R1 conformation and have led to widely used models for interpreting CW results of R1 in α‐helices.32, 34 By analogy, despite the non‐native side‐chain contacts present in the 15R1 and 44R1 crystal forms, the χ ₁—χ ₃ rotameric preferences are likely representative of fundamental R1 torsional preferences and are therefore of value.

Analysis of R1 rotamers in the crystal structures

A summary of all the rotamers observed for R1 at edge‐strand sites in the three crystal structures is given in Table 2. The table also contains a categorization of the χ ₁/χ ₂ values in the m (−60° ± 20°), p (60° ± 20°), t (180° ± 20°) notation.35 The preferred χ ₁/χ ₂ conformations of {m,m} and {t,p} have been observed previously for R1 at solvent‐exposed α‐helix sites.32, 33, 34 These observations support the possible existence of an intraresidue C_α—H_α···S_δ hydrogen bond that restricts R1 flexibility. The distance between the H_α and S_δ atoms are provided to probe the possible existence of this interaction in the new structures.

In the 15R1 mutant, the spin label resides at a non‐hydrogen‐bonding edge‐strand site where interactions with the neighboring chain are likely to disfavor formation of a restricting C_α—H_α···S_δ H‐bond.30 The data obtained here support that hypothesis. The χ ₁/χ ₂ conformations of four R1 rotamers observed in the structure of the 15R1 mutant match most closely with a {t,t} designation, while the remaining three resemble either {p,p} or {p,t} conformations. None of these conformations have been previously observed for R1 in a solvent‐exposed environment, though {p,p} has been noted in a membrane‐embedded β‐sheet context.36 Based on the χ ₁/χ ₂ conformational analysis as well as H_α—S_δ distances, none of the observed 15R1 rotamers has a stabilizing C_α—H_α···S_δ H‐bond. The existence of such an interaction is key to the analysis and interpretation of continuous wave ESR spectra in terms of backbone dynamics.11

The 44R1 mutant has the spin label at a hydrogen‐bonding edge‐strand site where local contacts with the remainder of the protein should be minimized and the C_α—H_α···S_δ H‐bond potentially able to form. One of the observed R1 rotamers in the two structures of this protein, 44R1_A1, is very similar to the {t,p} conformation expected to support such an interaction. Additionally, the χ ₃ value for 44R1_A1 is 72°, most similar to an expected value of χ ₃ of +90°. For the {t,p} conformation, a χ ₃ of +90° results in an R1 conformation that extends furthest away from the protein.34 The two remaining rotamers are best classified as {t,t} and {m,t}. The {t,t} conformation corresponds to the most commonly observed state in the 15R1 structure above, while {m,t} matches a previously observed structure of R1 at a solvent‐exposed internal β‐strand site.31

DEER distance measurements and comparison with crystal structures

The DEER experiment yields distance distributions between the unpaired electrons, which in this case are located within nitroxide groups of two R1 side chains. As detailed above, each R1 nitroxide ring is tethered to the protein through the five rotatable bonds. The dihedral angles about these bonds dictate the exact placement of the unpaired electron relative to the backbone (up to 7 Å away from the C_α), and fluctuation of these values can create an ensemble of different distances between the two spins. Thus R1 conformational behavior influences both the most probable distance observed by DEER as well as the distribution width.2 In order to use DEER as a means of understanding R1 behavior at edge‐strand sites, each modification described above (15R1, 44R1) was paired with an R1 spin label at the same solvent‐exposed α‐helical site (28R1). DEER experiments on the resulting double mutants (15R1/28R1, 28R1, 44R1) proved quite informative with respect to R1 conformational behavior.

Non‐hydrogen‐bonded site

The DEER distance distribution observed for the 15R1/28R1 mutant is displayed in Figure 4A. The most probable interspin distance is ∼25 Å, whereas the corresponding C_α—C_α distance between these sites in a crystal structure of wild‐type GB1 (PDB: 2LGI) is 16 Å. Thus, the R1 side chains add a total of 9 Å to the measurement. The distribution width (between 16 and 84% probability) of this DEER distribution is 7 Å.

Figure 4.

DEER results compared with MD (dashed) and crystal (vertical black lines) for 15R1/28R1 (A) and 28R1/44R1 (B). For the 28R1 α‐helical rotamers, χ ₃ was adjusted to −90° and the distances are shown in red. (C) The most commonly populated MD rotamers for each site. The {m,p} and {p,p} rotamers were not sampled at either site.

Previously we have shown that comparing experimental DEER distance distributions with predicted data based on R1 rotamers observed in crystal structures is a valuable means of exploring R1 conformational preferences.31 All crystal structures solved here display a native‐like fold of GB1, so all distance modeling was performed using crystallographically independent chains containing the various β‐sheet rotamers. As none of the crystallized mutants included the helix site, the R1 side chain there was modeled based on a previous published structure of 28R1.31 An important consideration in this analysis is the observed χ ₃. Both 28R1 rotamers exhibit χ ₃ values close to +90°; however, crystal structures of R1 in other contexts have found χ ₃ of −90°,33 and prior calculations suggest that χ ₃ values of ±90° are equally populated at α‐helical sites.28, 29 Therefore, guided by prior work,31 distances were also calculated after altering χ ₃ to −90° to appropriately represent R1 conformations in the α‐helical context.

The comparison of experimental and predicted DEER distance distributions for 15R1/28R1 is shown in Figure 4(A). The black line represents the average modeled distance and the gray bar represents the range of calculated distances depending on the R1 rotamers selected for the analysis. The experimental and theoretical distances for 15R1/28R1 show excellent agreement. The most probable distance predicted by the R1 rotamers observed in crystal structures is 24 Å, within 1 Å of the experimental most probable distance. This indicates that the R1 conformations observed in the crystal structure likely represent those populated in solution at the non‐hydrogen‐bonding site. Interestingly, when χ ₃ of the 28R1 rotamers is adjusted to −90°, the most probable distance shifts only slightly to 26 Å [Fig. 4(A)].

Hydrogen‐bonded site

In the case of 28R1/44R1, the most probable DEER distance [Fig. 4(B)] was found to be ∼ 24.5 Å, with a distribution width of 4 Å. The C_α—C_α distance between these sites in the wild‐type crystal structure of GB1 is 12 Å, so the R1 side chains contribute ∼12.5 Å to the observed distance (more than half the most probable distance).

Similar to the analysis described above for 15R1/28R1, comparison of experimental DEER data with distributions predicted based on crystal structures were also performed for 28R1/44R1 to explore R1 at the hydrogen‐bonding edge‐strand site. The results of the comparisons are shown in Figure 4(B). In contrast to the case for 15R1/28R1, the experimental data show poor agreement with the model. The most probable distance of ∼19 Å based on R1 rotamers in the crystal structure is 6 Å shorter than the experimental most probable distance. However, adjustment of the χ ₃ value at 28R1 to −90° leads to a drastic change in the average modeled distance of ∼23 Å. In particular, the case of 44R1_A1 is noteworthy because this rotamer leads to the two longest modeled distances of 24 and 25 Å. These distances match very well with the DEER most probable distance of 24.5 Å. Interestingly, 44R1_A1 is the rotamer that exhibits a conformation similar to {t,p} that spatially allows for the stabilizing C_α—H_α···S_δ H‐bond to form. This observation may indicate the interaction is more pronounced in solution than suggested by the crystal structures.

Molecular dynamics simulations

Molecular dynamics simulations have been used previously to account for R1 conformational preferences in order to interpret DEER measurements. Currently, the only available MD force fields for R1 were generated and validated for the residue at a solvent‐exposed α‐helix site.29 Importantly, the {χ ₁/χ ₂} conformational state of {m,m} observed within MD has been directly observed in crystal rotamers of R1 in solvent‐exposed α‐helices.31, 32, 33, 34 Nevertheless, sampling issues have been reported, particularly with transitions of χ ₃ between +90° and −90° about the disulfide bond.2, 37

The appropriateness of MD for modeling the two edge‐strand contexts investigated here was assessed. Thus, simulations were performed on the 15R1 and 44R1 mutants. As a means to provide direct comparison with DEER measurements, simulations were also performed on 28R1. Based on the hypothesis that the two R1 sites would behave independently in each double mutant, the ensemble of side‐chain conformations obtained from the three simulations of the single mutants were combined to generate computed nitroxide–nitroxide distance distributions comparable to the measurements obtained in the DEER experiments.

Non‐hydrogen‐bonded site

The distance distribution comparison for MD simulations and DEER measurements of 15R1/28R1 are shown in Figure 4(A). The most probable distance predicted by MD is remarkably similar (within ∼1 Å) to that observed by DEER. The distribution width by simulation is ∼50% narrower as compared to experiment, however there is substantial overlap. Interestingly, the observed {χ ₁/χ ₂} conformation of {t,t} from MD [Fig. 4(C)] matches closely with four of the seven rotamers observed by crystallography (Table 2). The {p,m} conformation was also observed in MD, a conformation which has been predicted previously using molecular modeling of R1 at a different non‐hydrogen‐bonding edge strand site. Taken together, the MD conformations observed match well with experiment and therefore may appropriately model R1 conformational behavior at non‐hydrogen‐bonded edge strand sites in β‐sheets.

Hydrogen‐bonded site

In contrast to the case of 15R1/28R1, the distributions for 28R1/44R1 [Fig. 4(B)] show poor agreement between simulation and experiment. The MD most probable distance is ∼5 Å shorter than observed in DEER and the majority of the simulation predicts much shorter distances than experiment. Interestingly, the conformations sampled within MD include the {m,m} and {t,p} conformations [Fig. 4(C)], the two expected conformations that allow for the C_α—H_α···S_δ H‐bond. Although {m,m} and {t,p} are expected given the hydrogen‐bonding location, it is likely that these observed conformations are not representative of what is occurring in solution due to poor agreement in the distance comparison. In contrast to the poor results from MD, it is noteworthy that the crystal rotamer which provides the best distance agreement is a {t,p}‐like rotamer, 44R1_A1. The essential difference between the crystal rotamer and the MD {t,p} rotamers is χ ₃. The 44R1_A1 crystal rotamer exhibits a χ ₃ of +90° while all of the MD {t,p} rotamers sample a χ ₃ close to −90°.

Alternate methods for distance modeling

A variety of computational tools have been developed to aid in structural interpretation of DEER distance distributions. Unlike the molecular dynamics simulations described above, these modeling algorithms require only coordinates for a protein scaffold on which to simulate R1 motion. Commonly utilized tools include MtsslWizard,20 a PyMol plugin that predicts R1 behavior through simple spatial sampling for conformations free from steric clashes, and MMM,19 a Matlab plugin that uses rotamer libraries generated from MD simulations of an isolated R1 residue in solution. Despite the fact that neither of these tools incorporates flexibility in the backbone or local side chains, both have been successfully used in conjunction with experimental DEER measurements to validate results and draw conclusions.38, 39, 40 Here, both tools were assessed for their ability to predict R1 behavior at solvent‐exposed edge‐strand environments of β‐sheets. The input model in both cases was the same wild‐type GB1 crystal structure (PDB: 2LGI). Additionally, MMM was used to sample R1 conformations at both cryogenic (175 K) and ambient (298 K) temperatures.

Non‐hydrogen‐bonding site

The experimental DEER distributions for 15R1/28R1 were compared with modeling results obtained by the three algorithms (MtsslWizard, MMM at 175 K, and MMM at 298 K), and the results are shown in Figure 5(A). The best agreement in most probable distance between prediction and experiment was found for MMM at 175 K (∼1 Å difference), followed by MMM at 298 K (∼3 Å difference), and MtsslWizard (∼5 Å difference). As 15R1 is a non‐hydrogen‐bonding edge site, interactions with the neighboring strand in the sheet are expected. The poor fit from MtsslWizard may be due to the tool treating the local protein environment as static. In solution, movement of nearby bulky side chains may allow for alternative R1 rotamers that are not sampled using the approach implemented in the program. Therefore, the resulting model and distance distribution may be biased by the specific protein structure selected as the input. MMM at 298 K samples a broader selection of R1 rotamers, and this results in a distribution with a most probable distance that is ∼3 Å too long. Interestingly, MMM at 175 K provides a very similar distribution as compared to that obtained from the MD simulations detailed above.

Figure 5.

DEER distance distributions compared with common, simple modeling techniques for both 15R1/28R1 (A) and 28R1/44R1 (B).

Hydrogen‐bonding site

Analysis of the data for 28R1/44R1 [Fig. 5(B)] shows similar agreement in most probable distance among the three methods: MMM at 175 K∼1 Å shorter, MMM at 298 K∼2 Å shorter, and MtsslWizard ∼2 Å longer. For 44R1, the purely sterics‐based model of MtsslWizard matches best with the DEER experiment in the distribution of distances, thought the most probable distance differs slightly. This solvent‐exposed hydrogen‐bonding site likely lacks the potential for steric bias, and MtsslWizard matches DEER better in this context as compared to the non‐hydrogen‐bonding site 15R1. Alternatively, while MMM also predicts most probable distances similar to DEER, the distributions display a significant population of shorter distances not observed in the DEER experiment.

Conclusions

The main goal of this study was to assess R1 rotameric preferences according to contextual differences between a non‐hydrogen‐bonding edge‐strand site and a hydrogen‐bonding edge‐strand site using a variety of methods. Though this work focused on edge strand sites of an antiparallel sheet, the conclusions drawn are likely consistent for parallel edge strand sites as well as the hydrogen‐bonded edge sites in both arrangements feature side chains protruding away from the neighboring sheet. The DEER results provide the experimental basis for comparison with the various techniques. Crystallography revealed R1 side‐chain rotamers at both types of edge‐strand sites that are distinct from R1 conformers observed previously in other secondary structure environments. This result underscores the fact that context matters with regards to R1 rotameric preferences. The DEER distance distributions are adequately rationalized by models based on the crystal structures, indicating that rotamers observed in the crystal lattices are relevant to R1 in solution. One rotamer for the hydrogen‐bonding site shows evidence for an intra‐residue C_α—H_α···S_δ H‐bond, though further experiments are required to test this hypothesis explicitly. In attempts to reproduce distances observed in the DEER experiments with common analysis techniques, no consistent trend was observed. While some good fits were obtained, rotameric preferences of R1 at edge‐strand sites is not consistently well reproduced using either MD or two common DEER modeling algorithms. These results illustrate the inadequacy of current modeling methods utilized for the interpretation of R1 DEER distance distributions within the edge‐strand environment. Accordingly, further work is necessary to fully understand R1 in these environments for edge‐strand sites to be reliably used for gaining accurate protein structural information from DEER results. Due to these complications, it is recommended to use caution when selecting R1 labeling sites within β‐sheets edge strand sites due a lack of accurate modeling tools. Use of alternative labeling methods that feature greatly reduced ambiguity would be an ideal means for overcoming these issues. In particular, recently developed rigid spin labels such as the bifunctional spin label RX,41 the double histidine Cu²⁺ chelating motif,8 and rigid Gd³⁺ tags42 would offer advantages in the β‐sheet environment.

Materials and Methods

Generation, expression, purification, and labeling of GB1 mutants

The mutant plasmids coding for E15C, E15C/K28C, T44C, and K28C/T44C GB1 were generated using previously described methods.8 The resulting plasmids were then transformed into BL21(DE3) E. coli cells and subsequently utilized for overexpression, purification, and eventual spin labeling via methods described previously.31

Crystallization, data collection, and structure determination

Crystals of the nitroxide‐labeled E15C and T44C mutants of GB1 were grown by hanging drop vapor diffusion. Solutions of protein in water varying in concentration between 20 and 30 mg/mL were mixed (1 + 1 μL) with the well solutions, and allowed to equilibrate at room temperature over wells containing crystallization buffer. A single crystal was solved for 15R1 (PDB ID: 5BMG), grown from a crystallization buffer composed of 0.1M magnesium chloride, 0.1M Tris pH 4.5, and 20% w/v PEG 4000. Two distinct crystals forms were obtained for the 44R1 mutant, denoted crystal form 44R1_A (PDB ID: 5BMI) and crystal form 44R1_B (PDB ID: 5BMH). The well solution for the 44R1_A crystal was 0.1M HEPES pH 7.5 and 1.4M sodium citrate. For the 44R1_B crystal, the well solution was 0.2M potassium sodium tartrate tetrahydrate, 0.2M sodium citrate pH 6.0, and 2M ammonium sulfate. Each crystal was flash frozen in liquid nitrogen after cryoprotection in well buffer supplemented with 25% v/v glycerol prior to diffraction analysis.

Diffraction data were collected using CuKα radiation on a Rigaku/MSC diffractometer with FR‐E generator. Data for 15R1 and 44R1_A crystals were obtained on a Saturn 944 CCD detector, while data for the 44R1_B crystal were obtained on an R‐AXIS HTC image plate detector. All diffraction experiments were carried out at a temperature of 100 K. Raw images were indexed, integrated, and scaled with d*TREK. Structure determination and refinement were carried out using the Phenix software suite. Structures were solved by molecular replacement using a published structure of wild‐type GB1 (PDB 2QMT) as the search model. In the case of 15R1, the data indexed as tetragonal but the actual space group was found to be P2₁ with pseudomerohedral twinning (twin law L, −K, H). Refinement for 15R1 made use of the twin refinement algorithm in Phenix. All data collection and refinement details are displayed in Table 1.

ESR measurements and analysis

All X‐band ESR experiments were performed on a Bruker Elexsys 580 spectrometer with a Bruker ER 4118X‐MD5 resonator or a Bruker ElexSys E680 X‐band FT/CW spectrometer with a Bruker EN4118X‐MD4 resonator. For the DEER experiment, the pulse sequence used was (π/2)ν₁–τ ₁–(π)ν₁–T–(π)ν₂–τ ₂–(π)ν ₁–τ ₂–echo.43 The pump frequency (ν ₂) was placed at the maximum of the nitroxide spectrum and the observer frequency (ν ₁) was offset at the maximum of the low field component (∼67 MHz downfield). The (π/2)ν₁ and (π)ν₁ pulses were set to lengths of 16 ns and 32 ns respectively while the (π)ν₂ pulse was set to 16 ns. τ ₁ and T were set to 200 and 160 ns respectively, and T was incremented by a stepsize of 10 ns for 128 points; τ ₂ was adjusted such that T + τ ₂ = 1300 ns. The raw time domain DEER spectrum was analyzed using DEER Analysis 2013.44 The temperature for all experiments was controlled with an Oxford ITC503 temperature controller and an Oxford ER 4118CF gas flow cryostat with liquid nitrogen as the primary cryogen.

Molecular dynamics simulations

The R1 spin labeled side was constructed at sites 15, 28, and 44 of a wild type (WT) structure of GB1 (PDB: 2LGI)45 in PyMol.46 R1 at each site were initially constructed to match rotamers resolved in the crystal structures from each site. The resultant structures were solvated in an explicit water box and counter ions were added to the box to neutralize the system. Molecular dynamics simulations were performed using NAMD47 with the CHARMM22 48 force field for the entire protein with the exception of R1, which was parameterized using force fields developed by Sezer et al.29 Energy of the system was initially minimized using a conjugate gradient method. The system was then heated to 300 K and equilibrated in an NPT ensemble for 1 ns at 1 atm via a Langevin piston. The backbone of the protein was restrained for all minimization and equilibration steps. Three 10 ns MD runs were performed for each site label site at steps of 1 fs. Convergence for all runs occurred by 2 ns and the final 8 ns of each run were analyzed to obtain the rotameric states of R1. The resulting dihedral angles were extracted with VEGA ZZ49 and visualized using PyMol. The MD simulations were performed separately and thus distance distributions were generated by modeling the MD frames at the appropriate locations on a static WT GB1 structure (PDB: 2LGI) and measuring 25,000 distances between randomly selected frames.