© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
Studying excited states of
proteins by NMR
spectroscopy
Frans A.A. Mulder1,2, Anthony Mittermaier1,2, Bin Hon3,
Frederick W. Dahlquist3 and Lewis E. Kay1
1Protein Engineering Network Centers of Excellence and Departments of
Medical Genetics, Biochemistry and Chemistry, University of Toronto,
Toronto, Ontario M5S 1A8, Canada. 2These authors contributed equally to
this work. 3Institute of Molecular Biology and Department of Chemistry,
University of Oregon, Eugene, Oregon 97403, USA.
Protein structure is inherently dynamic, with function often
predicated on excursions from low to higher energy conformations. For example, X-ray studies of a cavity mutant of T4
lysozyme, L99A, show that the cavity is sterically inaccessible to
ligand, yet the protein is able to bind substituted benzenes
rapidly. We have used novel relaxation dispersion NMR techniques to kinetically and thermodynamically characterize a
transition between a highly populated (97%, 25 °C) ground
state conformation and an excited state that is 2.0 kcal mol–1
higher in free energy. A temperature-dependent study of the
rates of interconversion between ground and excited states
allows the separation of the free energy change into enthalpic
(∆H = 7.1 kcal mol–1) and entropic (T∆S = 5.1 kcal mol–1, 25 °C)
components. The residues involved cluster about the cavity,
providing evidence that the excited state facilitates ligand entry.
T4 lysozyme is comprised of two domains1, with the L99A cavity (150 Å3) located in the C-terminal domain. Substituted ben-
zenes have been shown to bind to the L99A protein2 with rates
that vary from 325 s–1 (indole) to 800 s–1 (benzene)3.
Crystallographic4 and NMR5 data suggest that the intradomain
structure of the protein is little affected by formation of the cavity and that the conformations in solution and in the crystal are
very similar6. Thus, the rapid binding of ligands to the cavity is
possible only if conformational rearrangements from the predominant solution structure occur.
Here we show that the accessible conformations are populated
at levels of under a few percent, consistent with the observation
of only a single set of crosspeaks derived from the major conformer in 1H-15N correlation spectra. Obtaining information
about the ligand accessible state(s) from population-weighted
measures, such as chemical shifts, relaxation times, scalar or
dipolar couplings, is not possible. In contrast, Carr-PurcellMeiboom-Gill (CPMG)-based spin relaxation rates can be sensitive to the presence of minor conformers with populations as
low as 1%, as long as the rates of interconversion are on the ms
timescale and large chemical shift differences, δω, between the
states are present7.
Conformational exchange on a µs–ms timescale leads to a
modulation of the chemical shift of an NMR active nucleus,
resulting in a contribution to the effective decay of transverse
signal, termed Rex. This contribution is a direct result of the
chemical shift difference between the spin in each of the
exchanging conformations and can be suppressed by the application of radio frequency B1 fields7. Thus, the effective decay rate
constant of magnetization, R2eff, decreases as a function of
increasing B1 field strength, νCPMG. Relaxation dispersion profiles, R2eff versus νCPMG, can be measured and subsequently analyzed to obtain differences in chemical shifts between
exchanging sites, as well as kinetic (rates of interconversion) and
a
b
c
d
Fig. 1 Dispersion profiles of T4 lysozyme L99A. Profiles for Val 111 13Cγ2 at 1H frequencies of a, 600 MHz and b, 800 MHz, and Val 111 backbone
amide 15N spin obtained at c, 500 MHz and d, 800 MHz proton frequencies over a range of temperatures. Solid lines represent best-fit curves generated with global values of kex and pE (13C and 15N data grouped separately), and individually optimized values of δωC and δωN. By recording profiles
at a number of different static magnetic field strengths, it is possible to establish whether the exchange process is fast (kex >> δω), intermediate (kex
≈ δω) or slow (kex << δω) on the NMR chemical shift time scale24. All three cases have been observed in the present study. The dispersion profile of the
13Cγ2 of Val 111 increases with temperature, consistent with a process on the intermediate-to-slow time scale over the complete temperature range
examined. However, the exchange process for the amide 15N of Val 111 increases from slow to fast with temperature. The value of α = d(ln (Rex)) / d(ln
(B0)) (calculated using Eq. 14 of Millet et al.24 assuming an 800 MHz 1H frequency) ranges from 0.1 (10 °C) to 0.7 (25 °C) for Val 111 13Cγ2, and 0.3 (13
°C) to 1.6 (28 °C) for Val 111 15N. Experimental details are presented elsewhere10.
932
nature structural biology • volume 8 number 11 • november 2001
© 2001 Nature Publishing Group http://structbio.nature.com
letters
© 2001 Nature Publishing Group http://structbio.nature.com
Fig. 2 Values of the forward (kGE) and backward (kEG) rate constants and
the equilibrium constant (KG/E = kEG / kGE) obtained from global fits of kex
and pE to 15N (open circle) and 13C (open square) relaxation dispersion
measurements for all residues as a function of 1 / T, including best-fit
lines (see Methods). Inset, a schematic representation of ∆H, T∆S and ∆G
as functions of a reaction coordinate connecting the ground, transition
and excited states, respectively (calculated using conventional transition
state theory and T = 25 °C).
thermodynamic (populations) parameters describing the
exchange process8. Here we have used recently developed variable pulse spacing CPMG relaxation experiments9,10 to probe the
conformational exchange processes in L99A.
Relaxation dispersion NMR measurements
To generate as complete a picture as possible, relaxation dispersion profiles for a large set of spin probes, including backbone
amide nitrogens and side chain methyl carbons, were measured
on L99A samples that were labeled uniformly with 15N or selectively with 13C at methyl positions of Val, Leu, Ile (Cγ2 only), Ala
and Met residues. The selectively labeled sample was prepared
using 3-13C-pyruvate as the sole carbon source11,12. Methyl
groups are particularly good reporters of dynamics in this system because the cavity is hydrophobic and lined with a large
number of methyl-containing residues4. Over 150 backbone and
75 side chain sites are available for analysis using this labeling
scheme.
Dispersion profiles for the side chain 13Cγ2 (Fig. 1a,b) and
backbone amide 15N (Fig. 1c,d) spins of Val 111 have been
recorded at two different magnetic field strengths corresponding
to 1H frequencies of 600 and 800 MHz (13CH3 data) or 500 and
800 MHz (15N data). More than half of the profiles measured are
flat, indicating that many methyl 13C and amide 15N spins in the
protein do not sense an exchange process. The remaining 25–30
methyl and 50–60 amide profiles are retained for analysis. All of
these dispersion profiles are well fit (reduced χ2 of 1.2 and 2.2 for
13C and 15N profiles, respectively, on average) by a two-site
exchange process
kGE
G
E
kEG
where G and E correspond to ground and excited states, respectively, and kGE and kEG are the rate constants for the interconversion. Although in principle data can also be fit to more complex
models of exchange, the low χ2 values obtained using the twosite model suggest that such an analysis is not warranted. In general, distinguishing between two or three site exchanges, for
example, is not possible unless the rate of exchange between
pairs of sites differs by more than an order of magnitude13.
clustered within a relatively narrow distribution at each temperature, suggesting that a single dominant exchange process is present in L99A. For example, the values kex = 1030 ± 500 s–1 and
1370 ± 430 s–1 (at 25 °C) were obtained from fits of the dispersion profiles derived from backbone 15N and side chain 13C
methyl spins, respectively. Dispersion curves for all of the
amides at a given temperature were subsequently fit simultaneously with global values of pE, the fractional population of the
excited state, and kex; similar fits were performed with the
methyl data. The reduced χ2 values were low (χ2reduced,avg = 1.4
and 2.7 for methyl and amide data, respectively). Fits to the data
illustrated by the solid lines were obtained using the global
exchange parameters (Fig. 1).
The temperatures profiles of ln (kGE) and ln (kEG) were
obtained from fits of the relaxation dispersion curves (Fig. 2).
The rates obtained from the methyl-derived data (squares) agree
well with values extracted from the amide dispersion profiles
(circles), indicating that both probes are sensing the same
process. The value of the equilibrium constant, K = kEG / kGE, has
been obtained from the ratio of the exchange rates measured at
each temperature, and the temperature dependence of ln (K) has
been used to calculate ∆H = 7.1 ± 0.2 kcal mol–1 and ∆S = 17.2 ±
0.8 cal mol–1 K–1 for the G to E transition. These values show the
typical entropy-enthalpy compensation characteristic of an
order-disorder transition and are 10–20% of those estimated for
the complete unfolding of T4 lysoyzme under these conditions.
This argues that the excited state is more disordered than the
ground state and may be partially unfolded. A reaction coordinate describes the path from the ground state (G), through the
transition state (T) to the excited (E) state, with activation parameters derived on the basis of transition state theory (Fig. 2,
inset). A transition state model that consists of a smooth free
energy surface and a unique maximum along the reaction coordinate is indistinguishable over a small temperature range from
a model with a lower barrier and a rough enthalpic surface, as
modeled by the phenomenological Ferry Law14,15. To investigate
the range of parameters compatible with our data, we varied the
transition state enthalpy and extracted the values corresponding
to the standard deviation of enthalpy due to ruggedness, σH. A
value of σH = 2.5 kcal mol–1 is obtained for an enthalpic barrier
height of 0 for G to T and is in close agreement with
∼2–3 kcal mol–1, which was proposed as a “generic feature of collective protein dynamics involving many atoms and local interactions within a 5–10 Å radius of influence”15.
Structural basis of chemical shift modulation
The amide and methyl probes that show the largest contributions from exchange cluster in regions around the cavity in L99A
(Fig. 3). Specifically, the C-terminal half of helix E, helix F, the
EF and FG loops, as well as the hinge regions on either side of
helix I, show significant exchange broadening. Although successfully fitting models that account for the exchange contribuA single dominant exchange process
tion to linewidth at a given site in a protein is possible, obtaining
Initially, dispersion curves at each temperature were fit on a per physical insight into the process, which contributes to chemical
residue basis (see Methods). Notably, the kex = kEG + kGE values shift modulation in the first place, is more difficult. For example,
nature structural biology • volume 8 number 11 • november 2001
933
© 2001 Nature Publishing Group http://structbio.nature.com
© 2001 Nature Publishing Group http://structbio.nature.com
letters
Fig. 3 Slow dynamics cluster about the cavity. a, Structure of the T4
lysozyme L99A variant. b, Identical view, indicating the positions of
methyl groups (larger spheres) and backbone amides (smaller spheres)
included in the 25 °C analysis. Spheres are color-coded according to values of δω obtained from global fits of kex and pE, with white corresponding to no change in chemical shift — that is, no Rex contributions — and
dark blue corresponding to large chemical shift differences between
ground and excited states (>2 p.p.m. for 15N and >1 p.p.m. for 13C). Large
values of δω highlight significant structural changes between the G and
E states. The cavity created by the L99A mutation is indicated by the blue
wireframe.
modulation of the chemical shift of a residue may be the result of
motion of that residue or may reflect motion of a neighboring
group, such as an aromatic amino acid. Using Bovey-Johnson
theory16, ring current contributions to the chemical shifts of all
amide nitrogens and protons have been calculated and found to
be <1 p.p.m. in all cases. The 15N spins with the largest shift contributions from ring currents have completely flat dispersion
curves, indicating that, with the exception of 180° ring flips, the
majority of aromatics in the protein are rigid. In contrast to the
relatively small chemical shift contributions from ring current
shifts, δω values as large as 6–7 p.p.m. have been obtained from
fits of 15N dispersion curves, implying significant structural
rearrangements.
Experiments have also been performed that indicate that
methyl proton linewidths from only four residues, Ile 3, Met 102,
Met 106 and Ala 146, show contributions from exchange5. Since
contributions from ring current shifts affect 1H sites more than
13C, this indicates that only 13C dispersion profiles for these
residues may contain contributions from the motions of aromatics. In this regard, the large dispersion profiles of Met 102,
Met 106 and Ala 146 likely arise from local magnetic field fluctuations created by motion of Trp 138 residing on helix I, which is
dynamic (see above).
Ground state exchange with a disordered intermediate
The relaxation dispersion data are consistent with a process in
which L99A interconverts between a ligand-inaccessible ground
state and an excited, ligand-accessible state that is 2 kcal mol–1
higher in free energy. The excited state is characterized by a
decrease in the number of stabilizing interactions relative to the
ground state (7 kcal mol–1 increase in enthalpy) with an increase
in entropy, consistent with a more dynamic structure that allows
ligand binding. The G to E transition involves a conformational
rearrangement that includes the C-terminal portion of helix E,
helix F, helix I and the loops connecting these elements to the
protein. Notably, relaxation dispersion profiles at all 15N and 13C
sites in the wild type protein are flat, indicating that no interconversion occurs in the absence of the cavity. Moreover, the interconversion rates measured using relaxation dispersion (kGE =
20s–1 and kEG = 800 s–1 at 20 °C) are consistent with the ligand
exchange rates measured3, suggesting that ligand binding and
the dynamic process described above are coupled.
Partial unfolding to permit ligand entry is a common feature of
proteins with small hydrophobic targets. For uteroglobin, a PCB
and progesterone binding homodimer, ligand entry is facilitated
by the reduction of an interchain disulphide bond and partial
unfolding of the N- and C-termini17. The antigenic lipid presentation molecule, CD1b, is active at low pH, where two helices that
form a deep hydrophobic binding groove become partially denatured18. Order to disorder transitions of an α-helical binding site
cap in intestinal fatty acid binding protein are implicated as a
rate-limiting step in ligand association and dissociation19. Both
934
b
a
BmrR, a multidrug resistance activator, and a type II retinoic
acid-binding protein (R111M substitution) depend upon
unwinding and translation of α-helices to expose hydrophobic
binding sites20,21. Finally, a guanidinium-stabilized unfolding
intermediate of the estrogen receptor binds the hydrophobic fluorescent ligand ANS with high affinity and may resemble a binding-competent species accessed transiently under native
conditions22. Thus, the conformational exchange observed in T4
lysozyme L99A is likely to be a common feature of a large set of
ligand binding proteins. The methods presented here, therefore,
can be applied to a variety of biological systems and used to provide a quantitative kinetic and thermodynamic basis for understanding how ligands enter occluded binding sites and how low
populations of excited states might be implicated in function.
Methods
Relaxation dispersion analysis. Dispersion profiles for each 15N
or 13C site were individually fit to equations valid for fast23 or all8
exchange timescales describing the dependence of R2eff on νCPMG,
kex, pE and δω (see below) to yield residuals, χ2indiv. The choice of
equation was made based on the parameter α (ref. 24) with the fast
timescale equation used for α > 1.5 (Fig. 1). Subsequently, global
values of kex and pE were optimized using profiles from all 15N or 13C
sites with detectable Rex at a given temperature, with δω fit on a
per-site basis to yield a residual, χ2group, for each site. Because all
residues are fit collectively and a substantial fraction of the dispersion profiles at each temperature are derived from exchange
processes that are not fast on the NMR chemical shift time scale,
obtaining well-defined global values for kex and pE, as well as values
for δω for each residue (in the fast exchange limit only the product
pE(1–pE)δω2 and kex can be measured), is possible. The sites experiencing conformational fluctuations distinct from the global process
were identified by large values of χ2group / χ2indiv. Global values of kex
and pE were refit, excluding the site with the largest ratio. This procedure was repeated until the largest χ2group / χ2indiv ≤ 2, eliminating
∼15% of the data. Errors in kex and pE were obtained from the standard deviation of values from jacknife iterations in which a random
10% of data was excluded each time. The equation used to fit dispersions derived from residues in fast exchange on the NMR chemical shift timescale is23:
R2(νCPMG) = R2(νCPMG=∞) + (pGpEδω2 / kex) (1 – (4νCPMG / kex) tanh(kex / 4νCPMG)) (1)
where pG = (1 – pE), νCPMG = 1 / (4τ) and 2τ is the time between centers
of successive 180° pulses. All other dispersions were fit to8:
R2(νCPMG) = R2(νCPMG=∞) + kex / 2 – νCPMGcosh–1[D+cosh(η+) – D–cos(η–)] (2)
where D± = 1/2(±1 + (ψ + 2δω2) / (ψ2 + ζ2)1/2), η± = (±ψ + (ψ2+ζ2)1/2)1/2 /
2(2)1/2νCPMG, ψ = kex2 – δω2, ζ = –2δω(pGkex – pEkex) and it has been
assumed that the intrinsic transverse relaxation rates in the ground
and excited states are the same.
nature structural biology • volume 8 number 11 • november 2001
© 2001 Nature Publishing Group http://structbio.nature.com
© 2001 Nature Publishing Group http://structbio.nature.com
letters
Thermodynamic and kinetic analysis. Thermodynamic parameters were obtained by nonlinear least-squares fits to experimental
KG/E values using ln (K) = (∆S / R) – (∆H / (RT)), where R is the universal
gas constant. The values of ∆H and ∆S were used as constraints
when fitting conventional transition state theory parameters: ln (k)
= ln (kBT / h) + (∆S′ / R) – (∆H′ / (RT)), where kB is Boltzman’s constant,
h is Planck’s constant and primed parameters refer to the transition
barrier height in either direction (∆H = ∆H′GE – ∆H′EG). In all cases, 15N
and 13C data were fit together. Errors in the fit parameters were
estimated from jacknife simulations in which data from each temperature and nucleus type were excluded in turn for a total of 13
iterations.
Acknowledgments
This work was support by grants from the Medical Research Council of Canada
(L.E.K.) and the National Institutes of Health (F.W.D.). L.E.K. is a foreign
investigator of the Howard Hughes Medical Research Institute.
Correspondence should be addressed to L.E.K. email: [email protected]
Received 28 June, 2001; accepted 4 September, 2001.
1. Zhang, X.-J., Wozniak, J.A. & Matthews, B.W. J. Mol. Biol. 250, 527–552 (1995).
nature structural biology • volume 8 number 11 • november 2001
2. Morton, A. & Matthews, B.W. Biochemistry 34, 8576–8588 (1995).
3. Feher, V.A., Baldwin, E.P. & Dahlquist, F.W. Nature Struct. Biol. 3, 516–521 (1996).
4. Eriksson, A.E., Baase, W.A., Wozniak, J.A. & Matthews, B.W. Nature 355, 371–373
(1992).
5. Mulder, F.A.A., Hon, B., Muhandiram, D.R., Dahlquist, F.W. & Kay, L.E.
Biochemistry 39, 12614–12622 (2000).
6. Goto, N.K., Skrynnikov, N.R., Dahlquist, F.W. & Kay, L.E. J. Mol. Biol. 308, 745–764
(2001).
7. Palmer, A.G., Kroenke, C.D. & Loria, J.P. Methods Enzymol. 339, 204–238 (2001).
8. Carver, J.P. & Richards, R.E. J. Magn. Reson. 6, 89–105 (1972).
9. Loria, J.P., Rance, M. & Palmer, A.G. J. Am. Chem. Soc. 121, 2331–2332 (1999).
10. Skrynnikov, N.R., Mulder, F.A.A., Hon, B., Dahlquist, F.W. & Kay, L.E. J. Am. Chem.
Soc. 123, 4556–4566 (2001).
11. Rosen, M.K. et al. J. Mol. Biol. 263, 627–636 (1996).
12. Lee, A.L., Urbauer, J.L. & Wand, A.J. J. Biomol. NMR 9, 437–440 (1997).
13. Allerhand, A. & Thiele, E. J. Chem. Phys. 45, 902–916 (1966).
14. Ferry, J.D., Grandine, L.D. & Fitzgerald, E.R. J. Appl. Phys. 24, 911–921 (1953).
15. Denisov, V.P., Peters, J., Horlein, H.D. & Halle, B. Nature Struct. Biol. 3, 505–509
(1996).
16. Johnson, C.E. & Bovey, F.A. J. Chem. Phys. 29, 1012–1014 (1958).
17. Hard, T., Barnes, H.J., Larsson, C., Gustafsson, J.A. & Lund, J. Nature Struct. Biol. 2,
983–989 (1995).
18. Ernst, W.A. et al. Immunity 8, 331–340 (1998).
19. Cistola, D.P., Kim, K., Rogl, H. & Frieden, C. Biochemistry 35, 7559–7565 (1996).
20. Zheleznova, E.E., Markham, P.N., Neyfakh, A.A. & Brennan, R.G. Cell 96, 353–362
(1999).
21. Chen, X. et al. J. Mol. Biol. 278, 641–653 (1998).
22. Gee, A.C. & Katzenellenbogen, J.A. Mol. Endocr. 15, 421–428 (2001).
23. Luz, Z. & Meiboom, S. J. Chem. Phys. 39, 366–370 (1963).
24. Millet, O., Loria, J.P., Kroenke, C.D., Pons, M. & Palmer, A.G. J. Am. Chem. Soc.
122, 2867–2877 (2000).
935