10
MAGNETIC RESONANCE IN CHEMISTRY Magn. Reson. Chem. 2004; 42: 162–171 Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/mrc.1320 Structure determination of aligned samples of membrane proteins by NMR spectroscopy Alexander A. Nevzorov, Michael F. Mesleh and Stanley J. Opella Department of Chemistry and Biochemistry, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0307, USA Received 30 June 2003; Revised 1 August 2003; Accepted 1 August 2003 The paper briefly reviews the process of determining the structures of membrane proteins by NMR spectroscopy of aligned samples, describes the integration of recent developments in the interpretation of spectra of aligned proteins and illustrates the application of these methods to the trans-membrane helical domain of a protein. The emerging methods of interpreting the spectral parameters from aligned samples of isotopically labeled proteins provide opportunities for simultaneously assigning the spectra and determining the structures of the proteins, and also for comparing the results from solid-state NMR experiments on completely aligned samples with those of solution NMR experiments on weakly aligned samples. Copyright 2004 John Wiley & Sons, Ltd. KEYWORDS: NMR; PISEMA; dipolar couplings; micelle; bicelle; bilayer; PISA Wheel; Dipolar Waves; membrane protein INTRODUCTION The three-dimensional structures of membrane proteins are essential for understanding many biological functions and are of interest in structural genomics. Helical membrane proteins comprise one-third of the expressed proteins encoded in a typical genome. 1,2 . Other membrane proteins are primarily ˇ-sheet. Despite the importance and prevalence of helical mem- brane proteins, experimental determinations of their struc- tures remain problematic for the most widely used meth- ods, including x-ray crystallography and solution NMR spectroscopy. Although the difficulties are due to myriad technical issues, they have a common origin in the influ- ences (on the proteins and the samples) of the lipids that must be present for these proteins to adopt their functional structures. 3 As illustrated in Fig. 1, lipids can self-assemble in the presence of water to form micelles, bicelles or bilayers. As a result, a wide variety of samples of membrane pro- teins can be prepared, providing much-needed flexibility in the design of NMR experiments. 4 All aspects of NMR spec- troscopy are strongly influenced by molecular motions, and these considerations are paramount in studies of protein- containing lipid assemblies. Although proteins in micelles reorient rapidly enough to be studied with solution NMR Ł Correspondence to: Stanley J. Opella, Department of Chemistry and Biochemistry, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0307, USA. E-mail: [email protected] Contract/grant sponsor: National Institutes of Health; Contract/grant numbers: R37GM24266; R01GM29754; P01GM64676. Contract/grant sponsor: Biomedical Technology Resource for Solid-state NMR of Proteins; Contract/grant number: P41EB001966. methods, bilayer environments typically immobilize pro- teins on the timescales relevant to the operative nuclear spin interactions (milliseconds). 5–7 Solid-state NMR spectroscopy can be used to deter- mine the three-dimensional structures of membrane pro- teins in bilayer environments. The combination of intense radiofrequency irradiations and sample alignment results in high-resolution spectra with single-line resonances from each labeled site whose frequencies retain the orientational information inherent in the anisotropic chemical shift and dipole–dipole spin interactions. In this paper, we briefly review the process of determining the structures of mem- brane proteins by NMR spectroscopy, describe the integra- tion of recent developments in the interpretation of spectra of aligned proteins and illustrate the application of these methods to the transmembrane helical domain of a protein. Emerging methods of interpreting the spectral parameters from aligned samples of isotopically labeled proteins pro- vide opportunities for simultaneously assigning the spectra and determining the structures of the proteins, and also for comparing the results from solid-state NMR experiments on completely aligned samples and solution NMR experiments on weakly aligned samples. The dipole – dipole interaction is a robust source of spatial and angular information. 8 When the coupling is between two covalently bonded nuclei in an immobile aligned molecule, the interpretation is particularly straightforward. The inter-nuclear distance is fixed at the length of the chemical bond; therefore, the frequency splitting yields an accurate measurement of the angle between the bond and the direction of the applied magnetic field. These angles provide the principal input for structure determination as orientational constraints and are complemented by the angular dependence of the chemical shift frequency. Each resonance in a separated local field spectrum 9 obtained Copyright 2004 John Wiley & Sons, Ltd.

Structure determination of aligned samples of membrane proteins by NMR spectroscopy

Embed Size (px)

Citation preview

MAGNETIC RESONANCE IN CHEMISTRYMagn. Reson. Chem. 2004; 42: 162–171Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/mrc.1320

Structure determination of aligned samplesof membrane proteins by NMR spectroscopy

Alexander A. Nevzorov, Michael F. Mesleh and Stanley J. Opella∗

Department of Chemistry and Biochemistry, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0307, USA

Received 30 June 2003; Revised 1 August 2003; Accepted 1 August 2003

The paper briefly reviews the process of determining the structures of membrane proteins by NMRspectroscopy of aligned samples, describes the integration of recent developments in the interpretationof spectra of aligned proteins and illustrates the application of these methods to the trans-membranehelical domain of a protein. The emerging methods of interpreting the spectral parameters from alignedsamples of isotopically labeled proteins provide opportunities for simultaneously assigning the spectraand determining the structures of the proteins, and also for comparing the results from solid-state NMRexperiments on completely aligned samples with those of solution NMR experiments on weakly alignedsamples. Copyright 2004 John Wiley & Sons, Ltd.

KEYWORDS: NMR; PISEMA; dipolar couplings; micelle; bicelle; bilayer; PISA Wheel; Dipolar Waves; membrane protein

INTRODUCTION

The three-dimensional structures of membrane proteins areessential for understanding many biological functions andare of interest in structural genomics. Helical membraneproteins comprise one-third of the expressed proteinsencoded in a typical genome.1,2. Other membrane proteinsare primarily ˇ-sheet.

Despite the importance and prevalence of helical mem-brane proteins, experimental determinations of their struc-tures remain problematic for the most widely used meth-ods, including x-ray crystallography and solution NMRspectroscopy. Although the difficulties are due to myriadtechnical issues, they have a common origin in the influ-ences (on the proteins and the samples) of the lipids thatmust be present for these proteins to adopt their functionalstructures.3 As illustrated in Fig. 1, lipids can self-assemblein the presence of water to form micelles, bicelles or bilayers.As a result, a wide variety of samples of membrane pro-teins can be prepared, providing much-needed flexibility inthe design of NMR experiments.4 All aspects of NMR spec-troscopy are strongly influenced by molecular motions, andthese considerations are paramount in studies of protein-containing lipid assemblies. Although proteins in micellesreorient rapidly enough to be studied with solution NMR

ŁCorrespondence to: Stanley J. Opella, Department of Chemistryand Biochemistry, University of California, San Diego, 9500 GilmanDrive, La Jolla, California 92093-0307, USA.E-mail: [email protected]/grant sponsor: National Institutes of Health;Contract/grant numbers: R37GM24266; R01GM29754;P01GM64676.Contract/grant sponsor: Biomedical Technology Resource forSolid-state NMR of Proteins; Contract/grant number: P41EB001966.

methods, bilayer environments typically immobilize pro-teins on the timescales relevant to the operative nuclear spininteractions (milliseconds).5 – 7

Solid-state NMR spectroscopy can be used to deter-mine the three-dimensional structures of membrane pro-teins in bilayer environments. The combination of intenseradiofrequency irradiations and sample alignment resultsin high-resolution spectra with single-line resonances fromeach labeled site whose frequencies retain the orientationalinformation inherent in the anisotropic chemical shift anddipole–dipole spin interactions. In this paper, we brieflyreview the process of determining the structures of mem-brane proteins by NMR spectroscopy, describe the integra-tion of recent developments in the interpretation of spectraof aligned proteins and illustrate the application of thesemethods to the transmembrane helical domain of a protein.Emerging methods of interpreting the spectral parametersfrom aligned samples of isotopically labeled proteins pro-vide opportunities for simultaneously assigning the spectraand determining the structures of the proteins, and also forcomparing the results from solid-state NMR experiments oncompletely aligned samples and solution NMR experimentson weakly aligned samples.

The dipole–dipole interaction is a robust source of spatialand angular information.8 When the coupling is betweentwo covalently bonded nuclei in an immobile alignedmolecule, the interpretation is particularly straightforward.The inter-nuclear distance is fixed at the length of thechemical bond; therefore, the frequency splitting yields anaccurate measurement of the angle between the bond andthe direction of the applied magnetic field. These anglesprovide the principal input for structure determinationas orientational constraints and are complemented by theangular dependence of the chemical shift frequency. Eachresonance in a separated local field spectrum9 obtained

Copyright 2004 John Wiley & Sons, Ltd.

NMR of aligned membrane proteins 163

Figure 1. Representations of three common modes ofself-assembly for lipids in which short-chain lipids formspherical aggregates called micelles. A combination of long-and short-chain lipids forms bicelles where planar bilayershave their hydrophobic ends solvated by short-chain lipids toform a disk-like structure. Long-chain lipids naturally formextended planar bilayers.

with a high-resolution experiment, such as PISEMA10

or SAMMY,11 is characterized by the frequencies fromthe heteronuclear dipolar coupling and chemical shiftinteractions. Although higher dimensional experiments offerenhanced opportunities for spectral resolution of largerpolypeptides and additional frequencies as orientationalconstraints, the spectral patterns that result from thechemical shift and dipolar coupling frequencies aloneprovide adequate resolution for studies of small- andmedium-sized uniformly labeled polypeptides and indicesof regular secondary structure elements of ˛-helices12,13 andˇ-sheets.14

STRUCTURE DETERMINATION OFMEMBRANE PROTEINS

There are several distinct steps to structure determinationof membrane proteins, starting with the selection of targetsequences. Many of the preparative steps are similar tothose required for structure determination by other methods.However, particular attention needs to be paid to thedynamics of the lipid assemblies because of the wide-rangingeffects of global and local motions on NMR experiments.

Choice of polypeptideSince one-third of a typical genome consists of open readingframes corresponding to helical membrane proteins, thereare a vast number of potential targets, whether specificquestions are being asked about biochemical processes or thestructural characteristics of a proteome are being surveyed.A surprise revealed by the analysis of many genomes isthat the majority of helical membrane proteins are relativelysmall polypeptides with only one or a few transmembranehelices.1,2 Of course, there are also larger polypeptides in thiscategory, and the structure determination of the 300–400-residue proteins with seven trans-membrane helices is amajor goal of structural biology because of the key rolesof this class of proteins in signaling and binding of drugs.At present, many polypeptides with between 50 and 200residues are under investigation, and the experimentalmethods and instruments are being further developed sothat the structures of substantially larger polypeptides canbe determined.

Protein expressionExpression in bacteria enables relatively large amounts ofpolypeptides to be prepared with a variety of uniform andselective isotopic labeling patterns.4 A number of factors needto be considered in optimizing the expression system for thepreparation of samples for NMR studies. By synthesizingthe DNA encoding the polypeptide sequence, it is possibleto optimize the codon usage for E. coli. We have foundthe use of fusion proteins to be essential for the expressionof small membrane proteins in E. coli.15 However, othershave been able to express membrane proteins with His tagswithout additional added polypeptides.16 An important areaof research is the further development of expression systemsand the determination of optimal conditions for both smalland large membrane-associated polypeptides. This includesthe fusion system, the expression vector, the choice of cellsand all growth conditions, including the media, temperatureand induction conditions.

Protein isolation and purificationLarge amounts of the expressed fusion proteins can befound in the cell membranes, inclusion bodies or in afew examples in the cytoplasm. They are isolated throughstandard methods, often taking advantage of a His tagincluded as part of the fusion protein, and then cleaved fromthe fusion partner with cyanogen bromide, since the lipidsused for solubilization typically denature the proteases usedfor sequence specific cleavages. Alternative approaches areunder development so that methionine-containing sequencescan be studied. Purification is generally accomplishedwith HPLC; however, size-exclusion chromatography inthe presence of lipids is another option. After removal ofsolvents, the polypeptide is added to lipids and water, andthe mixture self-assembles to form the sample.

Sample preparationLipids are amphiphilic molecules with at least one longhydrocarbon chain and a polar or charged head group. Asillustrated in Fig. 1, they self-assemble to form three types ofsamples, and conditions can be found where all three yieldhigh-resolution NMR spectra of helical membrane proteins.17

Micelles are small and roughly spherical aggregatesof lipids with their hydrocarbon chains on the inside.They solubilize membrane proteins, and the lipid–proteincomplexes reorient rapidly enough to give isotropic spectra.5

Weak alignment can be induced in these samples through theuse of lanthanides18,19 or stressed polyacrylamide gels.20 – 22

Bicelles are prepared by mixing two different lipids, oneof which forms the extended bilayer portion and the otherforms the caps at the ends of the disks. The sizes of thebicelles can be adjusted through the ratios of the two typesof lipids, ranging from small ‘isotropic’ bicelles23 to largebicelles that are similar to extended bilayers.24

Bilayers (and large bicelles) are the most desirable lipidassemblies for structural studies of membrane proteins.They are very similar to biological membranes and provideexperimental conditions most likely to mimic accurately thefunctional environment of the proteins.3 Bilayers can bemechanically aligned between glass plates. The linewidths

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

164 A. A. Nevzorov, M. F. Mesleh and S. J. Opella

of the protein resonances demonstrate that the alignment iscomplete, with a dispersion similar to that observed in singlecrystals along the direction of the alignment.7,25 Bicelles canbe aligned magnetically ‘perpendicular’ to the direction tothe magnetic field, or ‘flipped’ by the addition of lanthanideions so that the disks have the same alignment as bilayers onglass plates.26

NMR experimentsThe lipids associated with the polypeptide are the source ofthe correlation time problems that plague NMR studies ofmembrane proteins. Since polypeptides in micelles reorientrelatively slowly, at least compared with soluble proteinsof similar size, micelle samples require the use of high-fieldNMR spectrometers and elevated temperatures. Even so,many of the most sophisticated and informative solutionNMR experiments cannot be applied to membrane proteinsin micelles because their short relaxation times result in theloss of signals during extended pulse sequences. The issueswith relaxation times also affect measurements of homonu-clear 1H/1H NOEs, and generally very few ‘long-range’NOEs can be observed and assigned in helical membraneproteins in micelles. As a result, the primary source of struc-tural information are residual dipolar couplings (RDCs) inweakly aligned samples.27,28

Solid-state NMR experiments on protein-containingbilayers and bicelle samples also benefit from high-fieldmagnets. The intense radiofrequency irradiations requiredfor line narrowing can cause sample heating. However,these and other experimental problems can be dealt with,and it is possible to obtain high-resolution two- andthree-dimensional spectra of completely aligned membraneproteins.7

The interpretation of the angular constraints determinedfrom the chemical shift and dipolar coupling frequenciesmeasured in weakly aligned micelle samples and in com-pletely aligned bilayer and bicelle samples provides theunifying theme for solution- and solid-state NMR studiesof membrane proteins.29 The spectral patterns generatedfrom the combinations of these frequencies serve to mapthe protein structures on to NMR spectra and are analyzedthrough the representations of PISA Wheels and DipolarWaves, structural fitting of spectra, and the calculation of thestructure from the orientationally dependent frequencies.

ANALYSIS OF NMR DATA FROM ALIGNEDSAMPLES

The angular constraints derived from solid-state NMR exper-iments on completely aligned samples provide reliableand precise structural information. The independent mea-surements of frequencies associated with the anisotropicheteronuclear dipole–dipole and chemical shift interactionsfor each backbone amide site in a protein relative to a singlereference frame, e.g. the magnetic field axis for completelyaligned lipid bilayers, means that errors do not propagate ina cumulative manner. This is an extremely important featureof the method that enables the combination of experimentalangular constraints from individual residues and the well-established covalent geometry (bond lengths, bond angles,

dihedral angles, planarity of the peptide linkages) of proteinsto be used as the basis for protein structure determinationwith atomic resolution.

While the initial motivation for the development of theinterpretation methods was the analysis of experimentalPISEMA spectra of completely aligned proteins,7,30 theemergence of residual dipolar couplings as a structuraltool allows the data obtained from weakly aligned solubleproteins27,28 or from membrane proteins in micelles18,19,22,31

to be analyzed in a parallel manner. We apply several inter-related approaches to the interpretation of the experimentaldata from aligned proteins. PISA Wheels12,13 and DipolarWaves20,32,33 involve the analysis of the periodic patternsobserved in the spectral data, and are applicable to resultsfrom both solid-state NMR of completed aligned samplesand solution NMR of weakly aligned samples. They areillustrated in Fig. 2. In the case of Dipolar Waves, theperiodic function is obtained by using a simple distributionon a cone assuming a constant angle between the N—Hbond and the helix axis. As a result, the dipolar couplingsof helices exhibit wave-like patterns as a function ofresidue number, and similar effects have been noted forthe chemical shift interaction.34 Dipolar Waves provideindependent validation for the geometry of ˛-helices. They

Figure 2. An ideal ˛-helix with dihedral angles D �65°, D �40° is tilted at 0, 30, 60 and 90° relative to the magneticfield axis, hence the lipid bilayer normal (A). This results in PISAWheels with different locations in a two-dimensional PISEMAspectrum (B). If the 1H–15N dipolar couplings are extractedfrom this spectrum, they oscillate with a periodicity of 3.6residues (C).

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

NMR of aligned membrane proteins 165

provide a direct and reliable measure of the regularity of˛-helices, since they are independent of the magnitudes andorientations of the 15N chemical shift tensor. Comparisonsof experimentally measured dipolar couplings,32 modelingstudies and bioinformatics35 have shown that the helicesfound in proteins typically satisfy this ideal approximationfairly well.

PISA Wheel patterns and their dependence on orientationare shown in Fig. 2 for four different tilt angles of an ideal ˛-helix. The corresponding Dipolar Waves, which represent thedipolar couplings as a function of residue number, are alsoshown. The observation of these patterns in experimentaldata enables the slant angles (tilts) of helices relative to thelipid bilayer normal to be determined. PISA Wheels have theadvantage that they do not require resonance assignments,although they are susceptible to distortions due to variationsin chemical shift tensors. By focusing on the dipolar couplingdimension, errors in the determination of tilt angles areminimized, and of course they are essentially eliminatedby analyzing Dipolar Waves. The sequential assignment of

one or the identification of a few residues by type and theirposition in the PISA Wheel (or the phase of the Dipolar Wave)directly determines the rotational orientation (polarity) ofthe ˛-helix about its long axis. Selectively labeled proteinsare readily prepared and have been instrumental in theidentification of the positions of residues in experimentalPISA Wheels and Dipolar Waves, and this is shown in thecase of a single label and also the patterns for multiplelabeled sites in Fig. 3. In addition to determining therotation of the helix in the bilayer, an extremely usefulaspect of this approach is the fact that the rest of thespectrum can be assigned based on the expected patternof assignments in a helical wheel, the so-called ‘shotgun’approach.36 The third approach is to utilize the frequenciesfrom the assigned resonances to calculate the structure, asdemonstrated since the inception of the method. ‘Structuralfitting’ is a recently developed37 complementary approachto complete structure determination. It has been shown toidentify subtle deviations from ideal helical structure andalso other features. This is accomplished by the fitting of the

Figure 3. For an ideal ˛-helix tilted at 20° relative to the magnetic field, the position of a particular residue in the wheel-like pattern isdetermined by its absolute rotation relative to the long axis of the helix. For multiple labels, the pattern of resonances is also uniquelya function of the rotation of the helix. Simulations are shown for a hypothetical Ala selective label and for a hypothetical Val label ofthe same helix.

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

166 A. A. Nevzorov, M. F. Mesleh and S. J. Opella

spectrum to a model structure assuming a constant peptideplane geometry. In the case where the resonances have beenassigned, the Ramachandran angles and are consideredto be the only degrees of freedom. The whole structureis assembled by the sequential walking from one residueto the next calculating the angles and directly fromtwo orientation-dependent frequencies for these residues.In principle, it provides an ‘assignment-free’ approach tostructure determination. In practice, at its present stage ofdevelopment, it can utilize partial assignment informationto determine protein structures in the absence of completesequential assignments.

Dipolar WavesThe frequency due to a dipolar coupling at a 15N-labeledbackbone amide site is determined by the angle � that anN—H bond forms with respect to the direction of the appliedmagnetic field:

��1H �15 N� D š�0�N�Hh4r3

N — H

3 cos2 � � 12

�1�

The N—H bonds are not colinear with the axis of a ˛-helix, and are tilted away by an angle υ (for an ideal ˛-helixυ D 15.8°). If the helix axis is tilted by an angle �T, one canadapt the expression for the distribution on a cone38 andwrite for residue n that

cos � D cos �T cos υ � sin �T sin υ cos(

2nT

C �0

)�2�

where �0 corresponds to the helix rotation about its axisand T is the helix periodicity, generally 3.6 for an ˛-helix.In combination, Eqns (1) and (2) are sufficient to describethe Dipolar Waves observed in solid-state NMR spectra ofcompletely aligned samples. In addition, the distribution ona cone can be generalized to describe the RDCs measured inweakly aligned samples:

��1H �15 N� D š�0�N�Hh4r3

N — H

Aa

ð(

3 cos2 � � 12

C 34

R sin2 � cos 2�

)�3�

where � also depends on the azimuthal rotation of the helixaxis �R about the z-axis of the alignment frame, and isgiven by:

ei� D ei�R

[sin �T cos υ C cos �T sin υ cos

ð(

2nT

C �0

)C i sin υ sin

(2n

TC �0

)] /sin ��4�

PISA WheelsTo be able to correlate the dipolar 1H–15N couplings with theassociated 15N chemical shift interaction, the two expressionsare usually written in terms of the orientation of the commonnth peptide plane, referred to as the molecular frame (MF),with respect to the external magnetic field (˛n, ˇn) (cf. Fig. 4):

�n�15N� D 11 sin2 ˇn sin2�˛n � �� C 22

ð cos2 ˇn C 33 sin2 ˇn cos2�˛n � �� �5a�

�n�1H �15 N� D š�0�N�Hh4r3

N — H

3 sin2 ˇn cos2 ˛n � 12

�5b�

Figure 4. Schematic drawing of two consecutive peptideplane orientations and the frames used for the relevant spininteractions. The orientations of external magnetic field B0

relative to the two peptides planes are given by the angles (˛n,ˇn) and (˛nC1, ˇnC1), respectively. The latter are related via arotational transformation involving the torsion angles n andn (see text).

Here � D 17° is the angle between the x-axis of the MFand the z-axis of the principal axis system (PAS) of the 15Nchemical shift tensor.39 The N—H bond length is takenas rN — H D 1.07 A. For all residues (except glycine andproline) the magnitudes of the principal values for the 15Nchemical shift tensor are typically taken to be 11 D 64 ppm, 22 D 77 ppm and 33 D 217 ppm. For glycine, 11 D 41 ppm, 22 D 64 ppm and 33 D 210 ppm.40 Proline residues aredealt with separately because of the absence of the amidehydrogen and their different chemical shift parameters.

The frequencies associated with two consecutive residuen and n C 1 can be related by the following recursiverelation connecting two adjacent peptide planes given by theorientations (˛n, ˇn) and (˛nC1, ˇnC1). It is most convenient towork in the irreducible spherical basis of rank 1:

YT�ˇnC1, ˛nC1� D YT�ˇn, ˛n�P�8n, 9n� �6�

Here the indexing of and takes into account thevariations in their values along the backbone. The vector Yis given in terms of un-normalized spherical harmonics by:

Y�ˇ, ˛� � Y�1�

1 �ˇ, ˛�Y�1�

0 �ˇ, ˛�Y�1�

�1�ˇ, ˛�

, Y�1�

0 �ˇ, ˛� D cos ˇ,

Y�1�š1�ˇ, ˛� D Ý sin ˇp

2eši˛ �7�

The transformation matrix P(, ) is given by a product oftwo Wigner rotation matrices41 of rank 1:

P�, � D D.1/�151.8°, , 109.47°�D.1/�0°, � � 180°, 34.9°��8�

In geometric terms, P(, ) brings the MF associated withthe nth peptide plane into coincidence with the molecularframe of the following residue, �n C 1�, as illustrated in Fig. 4.The first Euler angle in the first Wigner matrix of Eqn (8) isthe angle between the y-axis of the MF and the N—C˛ bondof the nth plane; the third Euler angle in the first Wigner

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

NMR of aligned membrane proteins 167

matrix is the tetrahedral angle (typically 110.5°); finally, thethird Euler angle in the second Wigner matrix is the anglebetween the C˛ —C bond and the y-axis of the MF of the(n C 1)th plane. The numerical values for the bond anglesreflect the geometry of a standard peptide plane42 and areassumed to be constant, regardless of the residue type orposition in the sequence. Setting D �65° and D �40°

in successive applications of Eqn (6) results in an ideal ˛-helix. The corresponding PISA Wheel is obtained by usingEqns (5a) and (5b) for the two relevant NMR frequencies.

Structural fittingClearly, the Ramachandran angles and may vary alongthe backbone and, when two-dimensional data are availablefor each residue, there is no need to consider them to beuniform along the polypeptide backbone. Indeed, if theorientation of the nth peptide plane, (˛n, ˇn), is knownthen the angles n and n can be found by using Eqn (6)followed by the application of Eqns (5a) and (5b) to the NMRfrequencies of the next residue, n C 1. However, owing tothe quadratic nature of Eqns 5(a) and (5b), there are multiplepossible solutions for n and n. Nevertheless, it is possibleto restrict the possibilities to nearly ˛-helical solutions in thecase of solid-state NMR spectra of ˛-helical domains. Forpractical reasons, it has proven to be more efficient to fit thefrequency points for residues n > 1 using a simplex searchfor n and n rather than to solve Eqns (5a) and (5b) directly.The simplex search makes it easier to take into account theexperimental error and also to narrow the range of solutionscorresponding to ˛-helical conformations.

In calculating a structure from solid-state NMR data,three sources of error can be anticipated. The first arisesfrom imperfections in the experiments, e.g. incompletedecoupling or sample misalignment, which can lead toincorrect measurements of the frequencies. The secondsource is the residue-to-residue variability of the magnitudesand orientations in the molecular frame of the principalcomponents of the chemical shift tensor. Finally, it is notalways clear whether deviations in the spectrum of a helicalprotein from an ideal PISA Wheel are due to the variations inthe Ramachandran angles and or to a combinationof the first two sources of error. Taken together, thesethree uncertainties determine the accuracy of the structuralfit since multiple solutions for and consistent withthe experimental measurement may be possible within thelimits of experimental errors and uncertainties in the spininteraction tensors.

To assess the error in structure calculations, we performa statistical analysis. For every calculation of the Ramachan-dran angles between two consecutive frequency points, eachof the principal values of the chemical shift tensor is allowedto vary within š5 ppm relative to their canonical values. Theexperimental accuracy for the determination of the spectralpositions is conservatively estimated to be š100 Hz in eachfrequency dimension; in other words, a solution for the tor-sion angles and is regarded as plausible if the calculatedfrequency point lies within a 100 Hz radius relative to theexperimental point. To pick up multiple torsion solutionsduring the fitting, the starting values for the angles and

are randomized within š10° relative to their ideal values,0 D �65°, 0 D �40°. The r.m.s.d.s of the multiple solutionsrelative to their average structure are then used to estimatethe accuracy and uniqueness of the structural fitting. Spectrahave been simulated for helices with idealized curves, kinks,unwinding and -bulges, and the structures back-calculatedin order to assess the sensitivity of this approach to structuralvariations.43 Taking into account all of the apparent sourcesof error, structures determined by direct calculation or struc-tural fitting reproduce the structural features of helices inproteins to within 1 A. Importantly, the overall features ofthe helices are reproduced with high fidelity.

Deviations from ideality have a pronounced effect on theappearance of PISA wheels and Dipolar Waves. In general,however, the 100° rotations between adjacent residues inthe sequence preserve the wheel-like pattern of resonancesfrom helical residues and the periodicity that is characteristicof Dipolar Waves. Without the influence of chemical shiftvariability, Dipolar Waves are highly predictable and reliableindicators of molecular structure.

EXAMPLE OF A MEMBRANE PROTEINSTRUCTURE

The 50-residue fd coat protein is a typical membraneprotein. We have used it extensively as a model systemfor the development of NMR methods. The protein hasa long hydrophobic transmembrane helix and a shorteramphipathic in-plane helix that are connected by a turnor loop, depending on whether it is in micelles or bilayers,and it has mobile C- and N-terminal residues. The secondarystructures and relative orientations of the helices in themembrane-bound form of fd coat protein can be directlydetermined from the experimental data and fits to sinusoidsshown in Fig. 5. The results of three experiments on twodifferent polypeptides, the full-length fd coat protein anda 20-residue peptide that corresponds to the N-terminalamphipathic helix of the coat protein, are analyzed in thefigure. The dipolar couplings in Fig. 5(A) were measured ona sample of the coat protein in completely aligned bilayers,whereas the residual dipolar couplings in Fig. 5(B) and (C)were measured from samples of the 50- and 20-residuepolypeptides, respectively, in weakly aligned micelles. Themicelle samples were weakly aligned in compressed gelsusing the method of Ishii et al.20 to generate measurable1H–15N RDCs. The protein has similar, but not identical,properties in bilayer and micelle environments. Based on theperiodicity of the oscillations of the dipolar couplings, thenumber of residues in the N-terminal amphipathic helix iswell defined and nearly identical in all three samples. Thelength and other properties of the hydrophobic helix in thefull-length protein are similar in micelles and bilayers. Theaverage error per measurement for the fit of a four-residuesliding window function is shown in Fig. 5(D), (E) and (F) andthe absolute phases for each window are shown in Fig. 5(G),(H) and (I). The rise in score between residues Q15 and I22in Fig. 5(D) and (E) is evidence of the lack of periodicity inthe structure of the residues in the loop connecting the twohelices. With the exception of the kink near residue 39, thehelices are straight within experimental error, as evidenced

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

168 A. A. Nevzorov, M. F. Mesleh and S. J. Opella

Figure 5. 1H–15N dipolar couplings are shown for the fd coat protein in oriented lipid bilayers (A), in SDS micelles (B) and for theN-terminal amphipathic peptide of fd coat protein (C). Scores corresponding to the deviations from an ideal 3.6 periodicity areshown in (D), (E) and (F) with the absolute phases of those sinusoids shown in (G), (H) and (I). The phases of these sinusoidscorrespond to absolute rotations of the helices that are illustrated by the helical wheel diagrams shown in (J) and (K) for the bilayerform and micelle form of the protein, respectively.

by the low fitting errors for each helix. The average errorper residue is 200 Hz for the solid-state NMR data shown inFig. 5(A) and 0.4 Hz for the solution NMR data for each helixin Fig. 5(B) and (C). These values are less than the linewidthsof the resonances in the spectra and the errors estimatedfor the measurements of the dipolar coupling frequencies inthese spectra.

The amphipathic helix begins at residue 7. It ends atresidue 19 in bilayers and at residue 17 in micelles, which is animportant difference. There are no discernible differences inthe N-terminal helix of fd coat protein owing to the presenceof the hydrophobic helix, demonstrating that the two helicesare independent structural entities. Moreover, the propertiesof this helix are the same in micelle and bilayer samples, withthe exception of the few residues near the C-terminus thatare reflected in the length of the helix, indicating that thishelix does not appear to be affected by curvature or otherproperties of the lipid assemblies. The positions of residuesF11, W26 and F42 are used to denote the rotations of thehelices in the context of helical wheel diagrams [Fig. 5(J),(K) and (L)]. This analysis can be performed independentof an explicit knowledge of the magnitude of the alignmentfor weakly aligned samples. Given values for the magnitudeand rhombicity of the alignment for weakly aligned samplesbased on the extreme values of the 1H–15N couplings,orientations for the helical segments can be calculated. Thetilt angles and rotations of the ˛-helices determined in their

alignment frames by Dipolar Waves are used to generatemodels of the fd coat protein that contain information aboutthe relative orientations of the helices.22

A detail of the membrane-bound form of the coat proteinstructure that may have structural significance when it isassembled into bacteriophage particles is the change in helixdirection near residue 39. This same kink is found in themembrane-bound form of the protein, in both micelles andbilayers, and in the structural form of the protein in virusparticles.44 The irregular patterns of the dipolar couplings ofthe residues connecting the amphipathic and hydrophobichelices demonstrate that there are substantial differencesbetween the tight bend in bilayers and the longer loopstructure in micelles. There is evidence from relaxation datathat these residues have internal mobility in the micellesamples.5 In bilayers, the trans-membrane helix begins atresidue 21, whereas in micelles this helix begins at residue26. Clearly the nature of the interface influences the bilayer instructural properties of membrane proteins. The small size ofthe interhelical loops in bilayer samples restricts the possiblerelative orientations of the two helices, thereby limiting theambiguities in helix orientation. This was taken advantageof in the complete structure determination of this protein,36

which has used both direct calculation of the structure fromthe individual frequencies and structural fitting.

The chemical shift and dipolar coupling frequenciesprovide the input for structure determination. First, the

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

NMR of aligned membrane proteins 169

spectrum of the membrane-bound form of fd coat proteinwas assigned using PISA Wheels. Then the structure wascalculated using Eqns (5) and (6) as described in theprevious section. The three-dimensional structure is shownin Fig. 6(A). Next, the structural fitting was performed ona partially assigned spectrum37 [Fig. 6(B)]. The fitting wasperformed based on the assignments of resonances byresidue type without assuming any specific order of thepeaks for that residue type (i.e. Ala, Val, Leu, Ile, Phe andGly). The only restriction is that all the residues participate ina nearly ˛-helical conformation, which significantly limits thepossible solutions for the torsion angles and and also thenumber of possible sequential assignments of the spectrum.The structures corresponding to such possible assignmentsand structural fits deviate from each other by less than 1 A.

Figure 7 represents a pictorial summary of the resultsfrom analyzing the patterns of Fig. 5 based on DipolarWaves and PISA Wheels. Here the overall structure has beendissected into three ideal ˛-helical domains corresponding tothe kinked transmembrane helix and the amphipathic helixas evinced by the fit to the Dipolar Waves. Notably, thestructures of the transmembrane domain shown in Fig. 6 areconsistent with the results shown in Fig. 7. They all show asimilar helix tilt of about 20° and a kink near residue 39.

The continuous structure of fd coat protein in lipidbilayers obtained from the ‘shotgun NMR’ approach36

to resonance assignment and structure determination issummarized in Fig. 8. It consists of three views, A, B andC, showing the relative position of the transmembrane andamphipathic domains and a top view of the transmembranehelix.

Figure 9 compares the structures of the transmembranehelical domain of fd coat protein in bilayers determined withthe three different methods of data analysis. The structure inFig. 9(A) consists of two separate idealized ˛-helical parts ofthe TM helix as obtained from the fit to the dipolar wave [fromFig. 5(A)]; Fig. 9(B) corresponds to the continuous structuralfit37 and Fig. 9(C) shows the structure obtained using theshotgun NMR approach.36 Two residues, Trp26 and Lys40,are labeled in each structure to show the helix twist. Whileall the three fits show a kink near residue 39 and a similaroverall helix tilt, the fit to the two ideal ˛-helical parts is

Figure 7. Ribbon representations of the structures asdetermined from Fig. 5. Ideal ˛-helical fragments are used toshow a kink in the transmembrane domain and theamphipathic helix. (A) Structure from solid-state NMR data;(B) structure in micelles; (C) amphipathic helix in micelles.

different from the fits of Fig. 9(B) and (C) whereas the r.m.s.d.between fits (B) and (C) is 1.17 A. The structure of Fig. 9(A) issomewhat longer than those of (B) and (C), and the positionof Trp26 in (A) is different from those obtained from both thestructural fit and the shotgun approach. These differencesmay be due to uncertainties in the values of the 15N tensorand bond angles when the chemical shift information isemployed in the fitting. On the other hand, the fit to thedipolar waves does not take into account changes in thetorsion angles and along the backbone, which may leadto propagation of error and the above structural variations.The differences observed among the methods for interpretingthe structural data are relatively minor and will require aconsensus on bond lengths and bond angles to resolve.

Figure 6. (A) The peaks corresponding to the transmembrane helix of fd coat protein in bilayers are assigned using PISA Wheelpatterns. The structure of this helix as calculated using those assignments is shown to form a helix that spans the lipid bilayer withan angle of 20°. (B) Fitting the spectrum by residue type (i.e. without assuming any specific order of peaks for the same residue type)results in an ensemble of similar structures that deviate from each other by about 1 A r.m.s.d.

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

170 A. A. Nevzorov, M. F. Mesleh and S. J. Opella

31 A

Figure 8. Complete structure determination of fd coat protein using assigned 15N chemical shift anisotropy and 1H–15N dipolarcouplings. The loop between the two helices is very short, allowing the relative orientations of the two helices to be well determined.

Figure 9. Structures of the transmembrane domain of themembrane-bound form of fd coat protein. (A) Ideal helicalresidues fit to Dipolar Waves; (B) average structure fromstructural fitting; (C) unique structure from direct calculation.

Future prospectsThe experimental methods and instruments used in NMRstructural studies of membrane proteins are still at an earlystage in their development, yet they are already powerfulenough to determine the structures of membrane proteinsbecause of the strong linkage between the spectral patternsand structure. The principal limitation to determiningrapidly the structures of many small, helical membraneproteins is the availability of isotopically labeled samples.Major efforts are under way to express membrane proteinsefficiently in E. coli. Even with purified polypeptides, thereare issues with the refolding and reconstitution into thevarious lipid environments that can take patience to workout. On the one hand, the use of solid-state NMR methodssolves the correlation time problem and removes limitson the molecular mass of the complexes; for all practicalpurpose, a typical membrane protein in lipid bilayersaligned between glass plates acts as if it were infinitelylarge. On the other hand, it is of interest to be able todetermine the structures of larger polypeptides, which havemore complex spectra because of the large number ofresonances. A goal for the next few years is to apply NMR

methods to proteins with seven transmembrane helices,and this requires further improvements in resolution andsensitivity. Progress is being made in implementing theseexperiments on magnets corresponding to 1H resonancefrequencies as high as 900 MHz, which will have a beneficialeffect on both aspects of the experiments. The spectroscopicexperiments are also being developed, especially for highfield applications. Triple-resonance on 13C and 15N labeledproteins will provide additional dimensions for resolutionand measurement of orientationally dependent frequencies.Moreover, they will provide access to information aboutside-chain and additional backbone sites of the proteins.All of these efforts are linked by the mapping of proteinstructures to the spectra of aligned samples by anisotropicnuclear spin interactions.

AcknowledgmentsThis research was supported by grants from the National Institutesof Health (R37GM24266, R01GM29754 and P01GM64676) and theBiomedical Technology Resource for Solid-state NMR of Proteins(P41EB001966).

REFERENCES

1. Wallin E, Von Heijne G. Protein Sci. 1998; 7: 1029.2. Ubarretxena-Belandia I, Engelman DM. Curr. Opin. Struct. Biol.

2001; 11: 370.3. Gennis RB. Biomembranes: Molecular Structure and Function.

Springer: New York, 1989.4. Opella SJ, Ma C, Marassi FM. Methods Enzymol. 2001; 339: 285.5. Almeida FC, Opella SJ. J. Mol. Biol. 1997; 270: 481.6. Howard KP, Opella SJ. J. Magn. Reson. B 1996; 112: 91.7. Marassi FM, Ramamoorthy A, Opella SJ. Proc. Natl. Acad. Sci.

USA 1997; 94: 8551.8. Pake GE. J. Chem. Phys. 1948; 16: 327.9. Hester RK, Ackerman JL, Neff BL, Waugh JS. Phys. Rev. Lett.

1976; 36: 1081.10. Wu CH, Ramamoorthy A, Opella SJ. J. Magn. Reson. A 1994; 109:

270.11. Nevzorov AA, Opella SJ. J. Magn. Reson. 2003; 164: 182.12. Marassi FM, Opella SJ. J. Magn. Reson. 2000; 144: 150.13. Wang J, Denny J, Tian C, Kim S, Mo Y, Kovacs F, Song Z,

Nishimura K, Gan Z, Fu R, Quine JR, Cross TA. J. Magn. Reson.2000; 144: 162.

14. Marassi FM. Biophys. J. 2001; 80: 994.

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171

NMR of aligned membrane proteins 171

15. Ma C, Marassi FM, Jones DH, Straus SK, Bour S, Strebel K,Schubert U, Oblatt-Montal M, Montal M, Opella SJ. Protein Sci.2002; 11: 546.

16. Wang J, Kim S, Kovacs F, Cross TA. Protein Sci. 2001; 10: 2241.17. Opella SJ. Nat. Struct. Biol. 1997; 4(Suppl. S): 845.18. Veglia G, Opella SJ. J. Am. Chem. Soc. 2000; 122: 11 733.19. Ma C, Opella SJ. J. Magn. Reson. 2000; 146: 381.20. Ishii Y, Markus MA, Tycko R. J. Biomol. NMR 2001; 21: 141.21. Chou JJ, Gaemers S, Howder B, Louis JM, Bax A. J. Biomol. NMR

2001; 21: 377.22. Mesleh MF, Lee S, Veglia G, Thiriot DS, Marassi FM, Opella SJ.

J. Am. Chem. Soc. 2003; 125: 8920.23. Glover KJ, Whiles JA, Wu G, Yu N, Deems R, Struppe JO,

Stark RE, Komives EA, Vold RR. Biophys. J. 2001; 81: 2163.24. Sanders CR, Landis GC. Biochemistry 1995; 34: 4030.25. Ramamoorthy A, Wu CH, Opella SJ. J. Magn. Reson. 1999; 140:

131.26. Prosser RS, Bryant H, Bryant RG, Vold RR. J. Magn. Reson. 1999;

141: 256.27. Bax A. Protein Sci. 2003; 12: 1.28. Prestegard JH, al-Hashimi HM, Tolman JR. Q. Rev. Biophys. 2000;

33: 371.29. Opella SJ, Nevzorov A, Mesleh MF, Marassi FM. Biochem. Cell.

Biol. 2002; 80: 597.

30. Opella SJ, Ma C, Marassi FM. Methods Enzymol. 2001; 339: 285.31. Chou JJ, Kaufman JD, Stahl SJ, Wingfield PT, Bax A. J. Am. Chem.

Soc. 2002; 124: 2450.32. Mesleh MF, Veglia G, DeSilva TM, Marassi FM, Opella SJ. J. Am.

Chem. Soc. 2002; 124: 4206.33. Mesleh MF, Opella SJ. J. Magn. Reson. 2003; 163: 288.34. Kovacs FA, Denny JK, Song Z, Quine JR, Cross TA. J. Mol. Biol.

2000; 295: 117.35. Walther D, Cohen FE. Acta Crystallogr, Sect. D 1999; 55: 506.36. Marassi FM, Opella SJ. Protein Sci. 2003; 12: 403.37. Nevzorov AA, Opella SJ. J. Magn. Reson. 2003; 160: 33.38. Ulrich AS, Watts A. Solid State Nucl. Magn. Reson. 1993; 2: 21.39. Wu CH, Ramamoorthy A, Gierasch LM, Opella SJ. J. Am. Chem.

Soc. 1995; 117: 6148.40. Oas TG, Hartzell CJ, Dahlquist W, Drobny GP. J. Am. Chem. Soc.

1987; 109: 5962.41. Arfken G. Mathematical Methods for Physicists. Academic Press:

Orlando, FL, 1985.42. Creighton TE. Proteins: Structures and Molecular Properties.

Freeman: San Francisco, 1993.43. Kim S, Cross TA. Biophys. J. 2002; 83: 2084.44. Zeri AC, Mesleh MF, Nevzorov AA, Opella SJ. Proc. Natl. Acad.

Sci USA 2003; 100: 6458.

Copyright 2004 John Wiley & Sons, Ltd. Magn. Reson. Chem. 2004; 42: 162–171