V O L U M E 2 6 N U M B E R 4
A P R I L 1 9 9 3
Registered in US. Patent and Trademark Office; Copyright 1993 by the American Chemical Society
Methodological Advances in Protein NMR AD BAX* AND STEPHAN GRZESIEK*
Laboratory of Chemical Physics, National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, Maryland 20892
Receiued August 13, I992
Since the first experimental observation of nuclear magnetic resonance (NMR) in bulk matter more than 45 years ago,lI2 its history has been punctuated by a series of revolutionary advances that have greatly expanded its horizons. Indeed, methodological and instrumental developments witnessed over the past two decades have turned NMR into the most diverse spectroscopic tool currently available. Applications vary from exploration of natural resources3 and medical imaging to determination of the three-dimensional structure of biologically important macromolecules.Pg The present Account focuses primarily on the meth- odological advances in this latter application, partic- ularly as they relate to the study of proteins in solution.
After Ernst and Anderson developed Fourier trans- form NMR" and demonstrated its use, the introduction of a second frequency dimension in NMR spectroscopy by Jeener in 197112 provided a critical trigger to the development of this field. An enormous variety of experimental schemes, all based on the two-dimensional (2D) concept and largely developed by the group of Ernst,l3 expanded the applicability of NMR to the characterization of quite complex molecules, including natural products, sugars, synthetic polymers, and
Ad Bax was born in the Netherlands. In 1981, he received hls Ph.D In applied physics from the Deltt University of Technology after conducting most of his graduate research with Ray Freeman In the Physical Chemlstry Laboratwy at Oxford Unhrersily, working on the development of two-dimensional NMR methods. Following a postdoctoral appointment at Colorado State University, working in solkl-state NMR, he Joined the National Insthutes of Health. where he presently is Chief of the Section on Biophysical NMR Spectroscopy. His work focuses on the development of improved NMR methods for the characterlzation of the structure and dynamics of biological macromolecules.
Stephan OIzesiek was born In West Germany and received his Ph.D in physics from the Free Universily of Berlin in 1988 for optical studles of proton release in bacteriomodopsin. After a venture in a small software company, he received a fellowship from the Roche Research Foundation for NMR studies of pharmaceutically relevant blomolecules. Part of his fellowship he spent at the NIH, where he presently is employed as a Vishlng Associate in the Section on Biophysical NMR Spectroscopy.
peptides. Nearly a decade ago, protein structure determination was added to the realm of applications by the introduction of new systematic procedures for spectral analysis, primarily developed by Wiithrich and co-~orkers .~ During the 19809, development of new experimental pulse schemes continued to increase the power and applicability of 2D NMR to structural characterization of biopolymers. The most important development was undoubtedly the addition of a third frequency dimension to the NMR spectra.14J5 The concept of 3D NMR is so similar to 2D NMR that no new formalism for the description of such experiments is required. The main problem that had to be solved for the development of such techniques was a way to record and process the enormous data matrices asso- ciated with such experiments. The second problem, as will be discussed later, was that sensitivity of the 3D experiments frequently is much lower than for anal- ogous 2D experiments unless the third dimension corresponds to the chemical shift of a 13C or 15N nucleus and isotopic enrichment is used.16J7 The advances in genetic engineering techniques that have occurred in the last decade enable many proteins to be overproduced
(1) Bloch, F.; Hansen, W. W.; Packard, M. Phys. Rev. 1946,69, 127. (2) Purcell, E. M.; Torrey, H. C.; Pound, R. V. Phys. Reu. 1946, 69,
(3) Jackson, J. A. Log Analyst 1984, 16-30. (4) Wiithrich, K. NMR of Proteins and Nucleic Acids; Wiley: New
(5) Clore, G. M.; Gronenborn, A. M. Crit. Rev. Biochem. Mol. Biol.
(6) Kessler, H.; Gehrke, M.; Griesinger, C. Angew. Chem., Int . Ed.
(7) Bax, A. Annu. Reo. Biochem. 1989,58, 223-256. (8) Wiithrich, K. Acc. Chem. Res. 1989,22, 36-44. (9) Kaptein, R.; Boelens, R.; Scheek, R.; van Gunsteren, W. F.
(10) Clore, G. M.; Gronenborn, A. M. Science, 1991,252, 1390-1399. (11) Ernst, R. R. Adv. M a p . Reson. 1966,2, 1-137. (12) Jeener, J. Ampere Summer School,BaskoPolje, Yugoslavia, 1971.
37-38.
York, 1986.
1989,24, 479-564.
Engl. 1988,27, 490-536.
Biochemistry 1988,27, 5389-5395.
This article not subject to U.S. Copyright. Published 1993 by the American Chemical Society
132 Acc. Chem. Res., Vol. 26, No. 4, 1993
and labeled in microorganisms with the NMR observ- able stable isotopes. With isotopic enrichment, sen- sitivity of many of the heteronuclear 3D experiments is sufficiently high to add yet another frequency dimension to the NMR spectrum, dispersing resonance frequencies in four orthogonal dimensions.1a20 In this Account, we will discuss the advantages and problems associated with extending the dimensionality of the NMR spectrum and will attempt to provide an answer to the question, How many dimensions do we really need?
Bax and Grzesiek
I. Principles of Multidimensional NMR Although the principles of 2D NMR have been
reviewed many times, we will briefly reiterate some of these in order to clarify the basis of 3D and 4D NMR. Most of the useful nD NMR experiments are of the so-called ucorrelated" type, in which the chemical shift of a nucleus is correlated with the chemical shifts of other nuclei based on an interaction between them. For example, in the important 2D NOESY experiment, protons are correlated on the basis of the dipole-dipole coupling between their magnetic moments, giving rise to magnetization transfer via the nuclear Overhauser effect (NOE). The pulse scheme used for the NOESY experiment is sketched in Figure la. In this scheme, three RF pulses are applied to the proton spins, and the scheme is repeated many times for systematically incremented durations of the time tl. The signals detected during the time t z are modulated by the frequencies present during the time tl, and a two- dimensional Fourier transformation of the acquired data matrix results in the 2D NMR spectrum.
In the NOESY spectrum, correlations between a resonance of proton A and proton B will be observed if A and B are sufficiently close in space (less than -5 A). However, before the distance information in the NOESY spectrum can be fully interpreted, it is nec- essary to assign each of the resonances in the lH NMR spectrum to ita site in the chemical structure. To accomplish this it is also necessary to record so-called J-correlated experiments, in which magnetization is transferred between chemically bonded nuclei via the J-coupling mechanism. The oldest 2D NMR pulse scheme, in which magnetization is transferred from one proton to another via 'H-lH J coupling, is known as the COSY experiment and is probably the most popular experiment in the NMR analysis of small molecules. This experiment requires that JHH be not much smaller than the 'H resonance line width. This line width is approximately proportional to the inverse of the
(13) Ernst, R. R.; Bodenhausen, G.; Wokaun, A. Principles of Nuclear Magnetic Resonance in One and Two Dimensions; Clarendon Press: Oxford, 1987.
(14) Vuister, G. W.; Boelens, R.; Kaptein, R. J. Magn. Reson. 1988,80,
(15) Oschkinat,H.; Griesinger,C.;Kraulis, P. J.;Ssrensen, 0. W.;Ernst, R. R.; Gronenborn, A. M.; Clore, G. M. Nature (London) 1988,332,374- 376.
(16) Fesik, S. W.; Zuiderweg, E. R. P. J. Magn. Reson. 1988, 78,588- 59%
176-185.
(17) Marion, D.; Kay, L. E.; Sparks, S. W.; Torchia, D. A.; Bax, A. J.
(18) Kay,L. E.; Clore, G.M.;Bax, A,; Gronenborn,A. M. Science 1990, Am. Chem. SOC. 1989,111, 1515.
249,411-414. (19) Clore, G. M.; Kay, L. E.; Bas, A,; Gronenborn, A. M. Biochemistry
1991,30, 12-18.
J. Am. Chem. SOC. 1991,113, 370-372. (20) Zuiderweg, E. R. P.; Petros, A. M.; Fesik, S. W.; Olejniczak, E. T.
90 90 n n n
DECOUPLE t2 15N I DECCWLE I Figure 1. Examples of timing diagrams of 2D and 3D NMR pulsesequences. (a) 2D NOESY experiment. (b) 20 'H-detected 1H-15N HSQC correlation experiment. (c) 3D pulse scheme for 3D 15N-separated NOESY-HSQC experiment, obtained by concatenating schemes a and b. Radiofrequency pulses are marked by vertical bars and have typical durations of tens of microseconds. Signal is acquired during the time tz (schemes a and b) and tS (c); each scheme is repeated many times while the duration of t l (and tz, for scheme c) is systematically incremented from 0 to -30 ms.
molecular tumbling rate and therefore increases ap- proximately linearly with the size of the protein. For larger proteins 'H-lH Jcouplings are frequently smaller than the line width, making the COSY experiment ineffective. Other J correlation techniques, such as the one depicted in Figure lb, can correlate the frequency of a proton with that of its directly attached heteroatom (13C or I5N). The heteronuclear one-bond couplings, ~ J C H (125-160 Hz) and 'JNH (-92 Hz), are much larger than 3JHH, and frequently as much as 50- 90% of the magnetization can be transferred from protons to their directly coupled heteronuclei.16 Con- sequently, such 2D heteronuclear shift correlation techniques are highly sensitive and can often be carried out even without isotopic enrichment.
The concept of 2D NMR is easily extended to higher dimensionality. For example, the two pulse schemes of Figure la,b can be concatenated in a manner depicted in Figure IC, yielding a 3D experiment. The signals, acquired during the time t 3 , are now obtained for many different tl and tz durations. As was the case in the 2D NOESY experiment, the data are modulated in the tl dimension by the frequencies of other nearby protons; however, in the t~ dimension the modulation frequency is that of the 15N nucleus that is directly attached to the observed proton. Consequently, a 3D Fourier transformation (with respect to the time variables tl, tz, and t 3 ) yields a 3D frequency domain NMR spectrum. For practical reasons, such a spectrum is usually displayed and analyzed as a series of adjacent cross sections through the 3D data matrix.
Figure 2 illustrates the practical advantage of 3D over 2D NMR. The region of the conventional 3D spectrum that displays the NOE interactions involving amide protons of the protein staphylococcal nuclease (156 residues) is shown in Figure 2b and exhibits a high degree of resonance overlap. Figure 2a represents a
F1
1-
2.
3,
4
5
6
7
8
9
10
Methodological Advances in Protein NMR
15N=121.4ppm
T 6 2 a 0
FEldr N 6 6 b
0 - 0 '
V 6 6 d G B 6 8 oA60
OT62 0 4 A94
096
K 7 0 0
V667:'
A W O
N 6 6 p
V66a
A69a 0
N 6 6 0
K70 1
t
Fl
1,
2
3
4
5
6
7
8
9
10
PPn
Acc. Chen. Res., Vol. 26, No. 4, 1993 133
*\.. - . . . - -A696
0'
c - K 7 0 0 t
9 0 * .
0 0
I I 1
9 8 7 F2
Figure 2. (a) One out of 64 parallel cross sections through the 3D 15N-separated NOESY spectrum of the protein staphylococcal nuclease, displaying NOE interactions involving amide protons that are attached to a 15N with a 121.4 f 0.3 ppm chemical shift. (b) Corresponding region of a regular 2D NOESY spectrum, displaying NOE interactions involving all amide protons. From ref 17.
slice taken through the 3D 15N-separated NOESY spectrum and displays NOE interactions only for amide protons attached to 15N nuclei with a chemical shift near 121.4 ppm. The entire 3D spectrum consists of 64 such slices, representing interactions to amides with 15N chemical shifts ranging from 130 to 105 ppm.
Although resonance overlap in the 3D spectrum is dramatically reduced compared to that in 2D, inter- pretation of the 3D NMR spectrum frequently is not necessarily straightforward. First, even if the chemical shift frequency is known for each proton in the protein, these shifts are often insufficiently unique to identify the proton. For example, if in the 3D NOESY spectrum we observe that the amide proton (at 6.7 ppm) of Ala- 69 (15N at 121.4 ppm) interacts with a proton at 4.2 ppm, this does not identify uniquely the second proton because nearly a dozen protons resonate in the 4.2 f 0.02 ppm region. Therefore, it is useful to disperse the 3D spectrum in yet another dimension in order to reveal the frequency of the 13C directly attached to the proton at 4.2 ppm. The pair of lH and 13C shifts associated with the second proton frequently identifies it in a unique or nearly unique manner, greatly facilitating the identification of NOE interactions. Extending the 3D experiment into four dimensions is straightforward and involves inserting a 2D lH-13C correlation scheme
in the 3D experiment of Figure 1c.l8 The power of 4D NMR is illustrated in Figure 3, for the amide proton of Asn-17 in the protein interferon-y, a homodimer with a total molecular weight of 31.4 kDa. Figure 3a shows a "strip" taken through the 3D spectrum at the F2 and F3 coordinates of the Asn-17 amide, and it shows the chemical shifts of all protons that have an NOE interaction with this amide proton. Figure 3b shows the corresponding slice through the 4D spectrum, and it yields both the 'H and 13C chemical shifts.
Clearly, 4D NMR presents a desirable and logical approach for identification of NOE interactions in larger proteins. The method is more "symmetrical" than the 3D l5N- or W-separated NOESY experiments, where the chemical shift of its attached heteronucleus is obtained for only one of the two interacting protons. The 4D method mentioned above is only suitable for separation of NOES between 15N- and 13C-attached protons. A conceptually similar, but technically more demanding, 4D experiment can separate NOE inter- actions between 13C-attached proton^.^^^^^ This is crucial for the unambiguous identification of interres- idue side chain-side chain NOE contacts.
Assignment Approach. For small proteins, as- signment of the lH spectrum can be made by combined analysis of NOE- and 'H-lH J-correlated 2D spectra.
134 Acc. Chem. Res., Vol. 26, No. 4, 1993 Bax and Grzesiek
Ll.0 0
...............................
................................................. 0- .e I - - Q
e
......................... 0 ......... .y .* ........... +E- *. ........................ I! e- 0
..... .................. N17
6 '3 F16
(c
N17 65 60 13C 55 50 PPm
Figure 3. (a) Strip from the 15N-separated 3D NOESY spectrum of interferon-y displaying the chemical shifts of aliphatic protons that have an NOE with the backbone amide of Asn-17. The strip is actually a narrow vertical band of a 2D cross section, such as shown in Figure 2a. (b) Cross section through the 4D 15N/13C- separated NOESY spectrum, displaying the chemical shifts of the protons that have an NOE interaction to the amide proton of Asn-17, together with the shifts of the 13C nuclei directly attached to these protons. Broken contours correspond to 13C nuclei in the 46-26 ppm chemical shift range, which have been aliased once in the 13C dimension. Adapted from ref 48.
However, as indicated earlier, 'H-lH J correlation techniques frequently fail for larger proteins because of the increased lH resonance line width. This problem is even worse for proteins enriched with 13C, because the I3C-lH dipolar interaction causes additional proton line broadening. As assignments cannot be made on the basis of NOE interactions alone, other experiments for obtaining through-bond J correlations are essential. Moreover, in order to utilize the additional 15N and 13C chemical shift information available from the 3D and 4D 15N- and W-separated NOESY spectra, it is necessary to assign all protonated 15N and 13C nuclei.
In recent years, a novel assignment procedure has been developed that is applicable to uniformly isoto- pically enriched pr~teins .~l-~* This procedure is quite different from the traditional approach and is based primarily on one-bond J couplings between adjacent atoms. The one-bond J couplings are relatively uniform and depend only weakly on conformation. Typical values for the relevant coupling constants are indicated in Figure 4. Equally important are the magnitudes of the transverse relaxation times, T2, which determine the resonance line widths. 2'2 depends approximately linearly on the molecular tumbling rate, i.e., on mo- lecular size, and inversely on the viscosity, but also on the degree of internal mobility and on local confor- mation. As a rough guide, residues that do not have a
(21) Oh, B. H.; Westler, W. M.; Derba, P.; Markley, J. L. Science 1988,
(22) Ikura, M.; Kay, L. E.; Bax, A. Biochemistry 1990,29,4659-4667. (23) Bax, A.; Ikura, M.; Kay, L. E.; Barbato, G.; Spera, S. Protein
Conformations; Ciba Foundation Symposium 161; Wiley: New York,
240,908-911.
1991; pp 108-135. (24) Clore, G. M.; Gronenborn, A. M. Prog. Nucl. Magn. Reson.
Spectrosc. 1991, 23, 43-92. ' (25) Marion, D.; Driscoll, P. C.; Kay, L. E.; Wingfield, P. T.; Bax, A.; Gronenborn, A. M.; Clore, G. M. Biochemistry 1989,28,6150-6156.
(26) Grzesiek, S.; Bax, A. J. Magn. Reson. 1992, 96, 432-440. (27) Palmer, A. G., III; Fairbrother, W. J.; Cavanagh, J.; Wright, P. E.;
Ranee, M. J. Biomol. NMR 1992,2, 103-108.
I H-C-H
b) I P I P
H-C-H _ .
H H H O
I I H-C-H I P H-C-H I P
d) -N
H I
I H-C-H
I O
Figure 4. (a) A dipeptide segment of a protein backbone with the approximate values for the J couplings which are essential for the assignment procedure in isotopically enriched proteins. ( b e ) Schematic diagrams of the nuclei that are correlated in the (b) HNCO, (c) HNCA, (d) HBHA(CBCACO)NH, and (e) CBCANH experiments. Nuclei for which the chemical shift is measured in the 3D experiment are marked by solid circles. Nuclei involved in the magnetization transfer pathway, but not observed, are marked by open circles. Magnetization transfer in these experiments is marked by curved solid lines, and the direction of the transfer is marked by arrows.
high degree of internal mobility in a globular protein of 20 kDa at 35 "C have line widths of -12 Hz for the amide proton, -7 Hz for the amide nitrogen in the 'H-coupled mode and -4 Hz in the 'H-decoupled mode, -15 Hz for 13C,, and -25 Hz for a I3C-attached lHa, The line width of the carbonyl carbon is dominated by chemical shift anisotropy and therefore proportional to the square of the applied magnetic field; for the 20- kDa protein, values of -6 Hz are observed at 500-MHz 'H frequency. Comparison of these numbers with the J values shown in Figure 4 indicates that for a protein of 20 kDa most one-bond J couplings are significantly larger than the line widths. This means that magne- tization can be transferred with high efficiency from one nucleus to its directly coupled neighbor. In this way a number of 3D J-correlated NMR experiments have been constructed, correlating the backbone and side-chain resonances in a manner schematically in- dicated in Figure 4. For example, the HBHA(CBCA- C0)NH experiment2* correlates the amide lH and 15N
(28) Grzesiek, S.; Bax, A. J. Am. Chem. SOC. 1992, 114, 6291-6293. (29) Grzesiek, S.; Bax, A. J. Magn. Reson. 1992, 99, 201-207. (30) Clubb, R. T.; Thanabal, V.; Wagner, G. J. Magn. Reson. 1992,97,
213-217.
Methodological Advances in Protein NMR Acc. Chem. Res., Vol. 26, No. 4, 1993 135
Table I. Experiments for Determining Backbone Assignments and Secondary Structure of Isotopically Labeled Proteins
labelinp" experiment purpose solvent 15N l3C S/N timePdavs ref
- 15N-separated HOHAHA H&, N(i), Ha(i) HzO + f 3 25 'SN-separated NOESY short range NOE HzO + 3 16,17 HNCO H N ( ~ ) , N ( ~ ) , CO(i-1) Hz0 + + ++ 2 22,26 HNCAb HN(~) , N(i), Cdi)/Ca(i-l) HzO + + + 2 22,26
-
HCACOb Hdi) , Ca(i), CO(i) D2O +/- + ++ 2 22,27 CBCA(C0)NH C&l)/Ca(i-l), N(i), H N ( ~ ) HzO + + + 2.5 28 HBHA(C0)NH H&-l)/Ha(i-l), N(i), H N ( ~ ) Hz0 + + + 2.5 28 CBCANH C ~ ( ~ ) / C ~ ( i ) / C ~ ( i - l ) / C a ( i - l ) , N(i), H N ( ~ ) HzO + + f 2.5 29 HN(CA)COb N(t), HN(~) , CO(i) HzO + + 2.5 30
Minimum measuring time, determined by the required digital resolution and the minimum number of phase-cycling steps. '++", "+", and "fr refer to the inherent sensitivity of the experiment. Experiments labeled ++ and + may be shortened significantly by the use of pulsed field gradient methodology. b For proteins with favorable resonance dispersion and sensitivity, these three experiments may not be necessary. c "+* indicates essential labeling, u-n indicates that no labeling should be used, and "+/-* indicates that labeling is not relevant.
resonances of one residue with the Ha and Hp chemical shifts of its preceding residue. Thus, for each amide two or three (in the case of nonequivalent HB protons) resonances are observed in the 3D HBHA(CBCAC0)- NH spectrum, and their coordinates in the Fl, Fz, and F3 dimensions of the 3D spectrum correspond to the lH, and 'HB (Fl), 15N (Fz), and ~ H N (F3) chemical shift frequencies.
Table I lists the experiments that are needed to make the backbone lH, 13C, and 15N assignments and to determine the secondary structure. Also indicated in this table is the approximate measuring time needed for acquiring each of these spectra. This measuring time should be considered to be only a rough estimate for proteins in the 15-25-kDa range at concentrations of ca. 1 mM in a volume of -0.4 mL. Longer measuring times may be needed for proteins that require high digital resolution because of particularly severe reso- nance overlap, for more dilute samples, or for proteins approaching the molecular size limit of a particular experiment, in which case the protein line width seriously degrades the sensitivity of the experiment. On the other hand, shorter measuring times could be afforded for spectra that are relatively well resolved and for sample concentrations significantly higher than 1 mM. In favorable cases, complete backbone assign- ments may be obtained using only a subset of the experiments listed in Table I. However, for proteins with substantial overlap in the Ha-C, correlation, as typically encountered for larger proteins rich in a-he- lices, the entire arsenal, including additional experi- ments not listed here, may be required.
For determining the tertiary structure of a protein, additional experiments need to be carried out for obtaining side chain resonance assignments and for collecting the required NOE data. For obtaining a "high-resolution" structure, even more experiments are typically needed to measure the numerous homo- and heteronuclear J couplings and to determine stereospe- cific assignments of nonequivalent methylene protons and methyl groups in valine and leucine residues. Consequently, approximately 45-90 days of measuring time is presently required to gather all spectral infor- mation needed for determining a high-resolution protein structure. For very soluble proteins (> - 2 mM) smaller than -20 kDa, it may be anticipated that recent advances in pulsed field gradient methodology may
shorten this m i n i m ~ m . ~ l - ~ ~ However, for experiments involving NOE or J-coupling measurements, which are usually limited by low signal-to-noise ratios, no dramatic improvement is expected. At present, analysis of the multitude of 3D and 4D spectra is even more time consuming than the data acquisition itself, particularly as software for completely automated analysis of the NMR spectra is not yet available. Development of such software, although conceptually quite straightforward, is frought with many practical problems that have not yet been solved in a general manner.
11. Sensitivity of Multidimensional NMR The sensitivity of 3D NMR experiments has been
treated in a rigorous manner by Griesinger et al.34 Here, we will briefly discuss some of the most critical factors. The sensitivity of multidimensional NMR experiments is determined primarily by the efficiency of the mag- netization transfer steps. For example, if in the NOESY experiment, sketeched in Figure la, only 0.1% of the nuclear spin magnetization of proton A is transferred to proton B, this immediately results in a 1000-fold reduction compared to the resonance intensity of proton A in a conventional single-pulse one-dimensional spec- trum. Another factor that affects the sensitivity of multidimensional NMR is the decay caused by relax- ation during the evolution periods of the experiment. For example, if for the NOESY experiment mentioned above high resolution in the final 2D spectrum is critical, it will be necessary to use relatively long acquisition times in the tl dimension, Le., the pulse scheme must be repeated for a range of tl durations that stretches from 0 to several times the transverse relaxation time, Tz, of the protons. Note that the experiments with long tl durations carry little signal and, therefore, decrease the signal-to-noise ratio obtainable per unit of time. Another important but frequently overlooked detail is the fact that each time the spectral dimen- sionality is increased, the sensitivity drops by 2'12 because both the real and imaginary component of the signal must be sampled in separate experiments. Consequently, 4D experiments, for example, are in-
(31) Vuister, G. W.; Boelens, R.; Kaptein, R.; Hurd, R. E.; John, B.;
(32) Davis, A. L.; Boelens, R.; Kaptein, R. J. Biomol. NMR 1992, 2,
(33) Bas, A.; Pochapsky, S. S. J. Magn. Reson. 1992, 99,638-643. (34) Griesinger, C.; Ssrensen, 0. W.; Ernst, R. R. J. Magn. Reson.
van Zijl, P. C. M. J. Am. Chem. SOC. 1991,113,9688-9690.
395-400.
1989,84,14-63.
136 Ace. Chem. Res., Vol. 26, No. 4, 1993
herently 2 times less sensitive than 2D experiments. An ingenious approach, applicable to a number of J-correlated experiments, which reduces this loss at the expense of an increase in the complexity of the experiment, has recently been proposed by Palmer et al.35
A0 indicated above, the efficiency of the magneti- zation transfer processes is the main factor determining the sensitivity. Consequently, many of the original homonuclear 3D experiments, which contain two rel- atively inefficient homonuclear magnetization transfer steps, require concentrated samples to overcome their low inherent sensitivity. Nevertheless, such 3D ex- periments can be extremely powerful in resolving ambiguities that are invariably present in 2D spectra of proteins in the 5-15-kDa molecular weight range.
How Many Dimensions Do We Need? With the recent introduction of a fourth dimension in NMR spectroscopy, the logical question to ask is what the optimal dimensionality of an NMR spectrum is. There is no single answer to this question, but the following discussion is intended to clarify some of the issues. As pointed out by S~rensen,3~ 3D or 4D NMR spectra can be considered as mathematical products of their cor- responding 2D building blocks. Consequently, the 3D or 4D experiments do not offer fundamentally new information but merely resolve overlap problems present in the 2D spectra. Thus, the main purpose of increasing the dimensionality of the NMR spectrum is to reduce spectral overlap. Alternatively, the resolution in the 2D spectrum may be increased for the same purpose by using longer acquisition times in the orthogonal time dimensions. However, as indicated above, extending the acquisition time much beyond the applicable transverse relaxation time Tz rapidly decreases the inherent sensitivity of the experiment, and the payoff in increased resolution associated with longer tl acquisition times drops sharply. Therefore, if acceptable spectral resolution cannot be obtained with tl acquisition times on the order of the applicable Tz, increasing the spectral dimensionality will be useful, provided that an efficient additional magnetization transfer step is available. For proteins isotopically enriched with 13C and/or 15N, such an additional transfer usually can be generated quite efficiently.
Once the spectrum is adequately resolved, it usually does not pay to increase the spectral dimensionality any further. Consider, for example, a J-correlated experiment in which we are trying to correlate the 13C, and 'Ha resonances of one residue with the 15N and ~ H N backbone amide resonances of the next residue, and suppose, for convenience, that the 2D I5N-lH correlation spectrum does not yield any spectral overlap. We now have the choice of correlating resonances of all four nuclei simultaneously in a 4D experiment or conducting two 3D experiments, one which correlates the amide resonances with 13Ca, and one which correlates them with IHa. To make the example less abstract, let us assume identical acquisition times in the 15N and 'HN dimensions for the 3D and 4D experiments, and let us also assume that 8 complex increments are taken in the 13Ca and in the lH, dimension of the 4D experiment,
(35) Palmer, A. G., 111; Cavanagh, J.; Byrd, R. A,; Rance, M. J. Magn.
(36) Smensen, 0. W. J. Magn. Reson. 1990, 89, 210-216. Reson. 1992,96,416-424.
Bax and Grzesiek
i.e., the 2D 15N-lH correlation experiment must be repeated (2 X 8) X (2 X 8) = 256 times. In the same amount of time, two 3D experiments could be recorded with 64 complex increments each in their respective 'H, and 13C, dimensions. In this case, the two 3D experiments determine the peak position of the lHa/ 13Ca pair up to 64 times more precisely than does the 4D experiment. This example merely serves to illustrate that the highest possible dimensionality is not neces- sarily always the best choice. On the other hand, as argued by Boucher et al.,37 there may be other practical reasons that make it preferable to obtain information from a single 4D experiment instead of from two separate 3D spectra that have been acquired at different times, possibly under different conditions. Clearly, there are advantages and disadvantages associated with increasing the dimensionality of the NMR spectrum. In practice, an increase in the spectral dimensionality requires that an additional, efficient magnetization transfer step be available. For example, in the 4D NOESY experiment, the two transfer steps, added to the 2D NOESY scheme, involve correlating the two protons with their directly attached heteroatoms. With few exceptions, however, increasing the dimensionality in non-NOESY type experiments is necessary and beneficial only if resolution in the 3D spectrum is limited by the natural resonance line width, and not by the use of acquisition times that are shorter than the transverse relaxation time, T2. If resolution is not limited by Tz, a combination of two or three experiments of the lower dimensionality frequently is more efficient at yielding the desired information than is an increase in dimen- sionality.
111. Structural Parameters To date, NMR protein structures have been calcu-
lated almost exclusively on the basis of interproton distances (derived from NOE measurements) and dihedral angles (derived from lH-IH J couplings). In addition, slow exchange of amide protons with water is usually interpreted as an indication of hydrogen bonding.
Over the past decade, complete lH resonance as- signments have been made for a large number of proteins for which the 3D structure is accurately known from X-ray crystallographic studies. Careful analysis of the 'H chemical shifts indicates that they can be predicted on the basis of the protein structure to within a few tenths of a part per million.38 Therefore these chemical shifts may become useful indicators for the accuracy of a protein structure, or alternatively, they potentially could be used for further structure refine- ment.
The introduction of uniform isotopic enrichment yields access to additional structural parameters that may increase further the level of detail at which both the structure and the internal dynamics of a protein can be studied. These parameters are briefly discussed below.
15N and I3C Chemical Shifts. The new isotope- assisted methodology (Table I) relies on 13C and l5N chemical shifts to resolve the very crowded lH spectra
(37) Boucher, W.; Laue, E. D.; Campbell-Burk, S.; Domaille, P. J. J.
(38) Osapay, K.; Case, D. A. J. Am. Chem. SOC. 1991,113,9436-9444. Am. Chem. SOC. 1992,14, 2262-2264.
Methodological Advances in Protein NMR
of larger proteins. The 13C and 15N resonances them- selves also contain valuable information, however. For example, a clear correlation has been found between the protein backbone angles (b and $ and the 13Ca and 13CB chemical ~ h i f t s . ~ ~ ? ~ ~ The 13Cg resonance follows the same trend as the Ha proton resonance: for extended structures, with (b, $ - 130°, on average a downfield shift from the random coil position is observed, whereas for a-helical structures (4, $ - -50°), a small upfield shift is observed. The 13C, resonance follows the opposite trend, with significant downfield shifts (- 3 ppm) for helical structures and an upfield shift (- 1-2 ppm) for extended structures.
The 15N chemical shifts of backbone amides in proteins can also deviate substantially from their random coil values. For example, the 15N shifts of Val- 39 and Val-99 in staphylococcal nuclease differ by 32 ppm. Clearly, such deviations must be related to local structural differences. On average, amides in a P-sheet resonate downfield by about 5 ppm compared with a-helical amides, but large exceptions to this rule occur’. Attempts to correlate structural features with 15N chemical shifts a t a more detailed level so far have remained unsuccessful. Likely, there are a number of important contributors, including the (b and + backbone angles, the planarity of the peptide bond, and hydrogen bonding of the amide proton and its adjacent carbonyl.
JCouplings. To date, use of J couplings in protein structure determination has been restricted mainly to the three-bond H r H , and Ha-H~ couplings. These couplings are correlated with dihedral angles following well-known Karplus equations and have been used to restrain the (b and x1 angles, and for making stereospe- cific assignments of nonequivalent Ho methylene res- onances. In larger proteins, where these couplings typically cannot be obtained from the ‘H-lH multiplet structures, they may be measured using a variety of different methods that utilize the presence of stable isotopes. These isotopes also provide access to a large number of heteronuclear J couplings, and the three- bond lH-15N and lH-13C couplings carry particularly important conformational information that can be interpreted readily using Karplus equations.
Although little used to date, one-bond lH-13C J couplings are also related to structure in a simple manner, and these couplings can readily be measured in isotopically enriched proteins.41 Quite recently, a technique has been described for measuring long-range 13C-13C J couplings in proteins.42 For proteins in the 15-20-kDarange, these relatively small couplings (< -4 Hz) can be measured only for methyl carbons which, because of their nonexponential transverse relaxation, have a narrow component to their resonance line shape. These J couplings provide very direct information on the side-chain torsion angles in Leu, Ile, Val, and Thr residues.
I6N and lSC Relaxation Times. The degree of internal protein flexibility not only is of fundamental interest but also is critical for understanding protein recognition. For proteins at natural isotopic abundance,
(39) Spera, S.; Bax, A. J. Am. Chem. SOC. 1991, 113, 5490-5492. (40) Wishart, D. S.; Sykes, B. D.; Richards, F. M. J. Mol. Biol. 1991,
(41) Vuister, G. W.; Delaglio, F.; Bax, A. J. Am. Chem. SOC. 1992,114,
(42) Bax, A.;Max, D.; Zax, D. J. Am. Chem. SOC. 1992,114,69244925,
222,311-333.
9674-9675.
Acc. Chem. Res., Vol. 26, No. 4, 1993 137
the dynamic behavior of a protein is not easily quan- tified by NMR. In contrast, for proteins isotopically enriched with 15N and/or 13C, this information can be retrieved from 15N and 13C longitudinal and transverse relaxation times and from the heteronuclear 15N{lHJ and 13C(lHJ N O E S . ~ ~ The strength of this approach for the characterization of dynamics stems from the fact that the 15N and 13C relaxation times are dominated by the time dependence of the strong heteronuclear dipolar interaction with their attached proton(s). This time dependence is determined primarily by reorien- tation of the internuclear bond vector, and the relax- ation measurements provide information on both the amplitudes of individual angular bond vector fluctu- ations and the time scale on which they occur. For practical reasons, most of the detailed protein relaxation measurements so far have focused on backbone amides. For example, we used this methodology to show that the so-called “central helix” of the protein calmodulin, which separates its two globular domains in the crystalline state, functions as a flexible linker in solution.46 Specifically, the amide N-H bond vectors of four adjacent residues near the middle of this central helix show large angular fluctuations on a 100-ps time scale, and the anisotropy of the overall molecular tumbling, expected for a rigid “central helix”, was not observed. In contrast, each of the globular domains reoriented in a nearly isotropic manner, with the time scale of the smaller of the two domains being faster than that of the larger domain.
Extending the relaxation measurements to aliphatic side chain carbons in principle is straightforward. However, in practice extra care needs to be paid to technical details, related to 13C-13C J couplings, which can affect measurements of l3C transverse relaxation times and to analysis of the data when more than one proton is coupled to a particular 13C nucleus. In the latter case, dipolar cross correlation can affect the quantitative interpretation of measured parameters. Although the theoretical formalism for describing this effect is well established, for proteins cross correlation can seriously affect the measured parameters them- selves unless specific precautions are takena47
IV. Conclusion and Outlook
It is clear that the introduction of 3D and 4D NMR, combinedwith uniform 13C and 15N isotopic enrichment, significantly extends the molecular weight limit of proteins for which a solution structure can be deter- mined by NMR. However, it is equally clear that such structural studies of larger proteins are time-consuming and costly. The minimum time for data acquisition needed for detailed characterization of a single protein in the 25-35-kDa range is expected to remain at least several months. Tens of milligrams of isotopically enriched protein, which must remain stable over such long measuring periods, are required for this process.
(43) Lipari, G.; Szabo, A. J. Am. Chem. SOC. 1982, 104,4546-4558. (44) Dellwo, M. J.; Wand, A. J. J. Am. Chem. SOC. 1989,111, 4571-
(45) Kay, L. E.; Torchia, D. A,; Bax, A. Biochemistry 1989,28,8972-
(46)Barbato, G.; Ikura, M.; Kay, L. E.; Pastor, R. W.; Bax, A.
(47) Kay, L. E.; Bull, T. E.; Nicholson, L. K.; Griesinger, C.; Schwalbe,
4518.
8979.
Biochemistry 1992,31, 5269-5278.
H.; Bax, A.; Torchia, D. A. J. Magn. Reson. 1992, 100, 538-558.
138 Acc. Chem. Res., Vol. 26, No. 4, 1993
Depending on the efficiency and the type of expression system, this can require large amounts of expensive isotopically enriched precursors. Analysis of the mul- titude of 3D and 4D NMR spectra, needed for a detailed structural characterization, presently also is a labor- intensive and time-consuming process. However, de- velopment of suitable support software is expected to alleviate this tedious burden in the near future, significantly reducing the time needed for structure determination.
The presently well established 2D methodology for determining protein structure starts to fail when the lH line width, which is proportional to the protein molecular weight and inversely proportional to viscosity, becomes significantly larger than the homonuclear 'H- lH J couplings. For such larger proteins, resonance overlap in the 2D NOESY spectrum presents an additional serious barrier for structure determination. However, this latter limitation is less fundamental in nature since it may be resolved by going to higher field strengths or by recording homonuclear 3D experi- ments.14J5
For proteins enriched uniformly with 13C and 15N, the multinuclear methodology described in this Account works very well for intermediate size proteins but it starts to fail when the I3C line width becomes signif- icantly larger than the 13C-13C couplings. At room temperature this occurs for proteins of -30 kDa. Indeed, for the protein interferon-y (31.4 kDa), line widths of -45 Hz for the 13C, resonances are observed at 27 "C, making complete assignments of all side-chain resonances extremely difficult.48 Procedures for as- signment of the backbone resonances are, in our experience, more robust and function quite well, even for proteins as large as interferon?. Recording of the 4D NOESY spectra also becomes problematic for these larger proteins because the sensitivity loss occurring during the lH-W correlation step(s) in these experi- ments is significant. For example, in interferon-y the lH line widths of many of the 13C attached protons are
(48) Grzesiek, S.; Dobeli, H.; Gentz, R.; Garotta, G.; Labhardt, A. M.; Bax, A. Biochemistry 1992,31, 8180-8190.
Bax and Grzesiek
in excess of 50 Hz, resulting in a 3-fold loss in NOE cross peak sensitivity from relaxation during the 'H- 13C correlation step. Altogether, it therefore appears that 30 kDa presents an upper molecular weight limit for proteins that can be studied in detail with the technology outlined herein. Slightly larger proteins may be accessible if solvent viscosity can be decreased by raising the temperature or if the protein can be studied at concentrations significantly higher than 1 mM.
For completeness, it should be mentioned that the 13C- and 15N-based multidimensional approach de- scribed here is not the only possible approach for increasing the molecular weight limit of proteins that can be studied in detail by solution NMR. One powerful alternative approach utilizes either random or residue- specific deuteration, thus simplifying the lH spectrum and narrowing 'H resonance line ~ i d t h s . ~ ~ , ~ ~ This approach has the advantage that the required NMR experiments are all relatively simple and of the homo- nuclear 'H type. However, when residue-specific la- beling is called for, a large number of samples with different deuterated amino acid residues are needed, and obtaining a comprehensive set of resonance as- signments remains a difficult and labor-intensive task. It is conceivable that partial 2H labeling in combination with 13C- and 15N-based multidimensional experiments may further increase the molecular weight limit of proteins whose solution structure can be determined by NMR. Alternatively, a combination of the 13C- and 15N-based multidimensional experiments and residue- specific 13C- and/or l5N-labeling may prove to be effective for this purpose.
We thank Ted Becker, Marius Clore, Angela Gronenborn, David Live, John Schwab, Attila Szabo, and Dennis Torchia for stimulating discussions and useful suggestions during the preparation of this manuscript. The work in the authors' laboratory was supported in part by the Intramural AIDS Directed Anti- Viral Program of the Office of the Director of the National Institutes of Health. I so. (49) LeMaster, D. M.; Richards, F. M. Biochemistry 1988, 27, 142-
(50) Arrowsmith,C. H.;Pachter,R.;Altman,R.B.;Iyer,S.B.; Jardetzky, 0. Biochemistry 1990,29,6332-6341.