NMR supplement
504 nature structural biology ? NMR supplement ? july 1998
In a cell, the starting point of protein
folding is the nascent chain as it forms
on the ribosome. The process of protein
folding continues in a crowded molecu-
lar environment, in the presence of a
variety of helper molecules, the most
famous of which are the molecular chap-
erones whose major functions include
the control of protein aggregation. Many
small proteins, however, will refold effi-
ciently in dilute aqueous solutions fol-
lowing transfer from a denaturing
environment (such as 6 M guanidinium
chloride) into one where the native state
is thermodynamically stable (such as
that produced by dilution of the denatu-
rant). This allows the use of biophysical
techniques to follow the folding process
in real time. Two aspects of folding, how-
ever, make this task challenging. The
first is that folding is usually fast; many
small proteins fold in milliseconds or
less, although others may take consider-
ably longer. The second, and perhaps
most significant, is that the initial state
from which the folding reaction is initi-
ated is extremely heterogeneous
1
. The
ultimate starting point of folding is a
random coil, and proteins in strong
denaturants approach this rather close-
ly
2
. In the random coil state there are
more accessible conformations for a
polypeptide than there are molecules in
the test tube. This means that the process
of folding can involve extremely diverse
structural ensembles until the very last
stages of the reaction. This complicates
substantially the analysis of the results of
structural studies.
In order to combat these problems,
one approach has been to utilize a wide
range of spectroscopic techniques, each
able to monitor the formation of specific
aspects of protein structure, in stopped
and quenched flow mode
3
. NMR spec-
troscopy has played a significant role
here through its ability to analyze the
distributions of hydrogen and deuterium
in labile sites in proteins, and through
‘pulse labeling’ to follow in a site-specific
manner the formation of structure that
protects against solvent exchange, for
example as a result of the formation of
hydrogen bonds involving amide hydro-
gens
4
. Much has been learned through
these approaches, but recently, increas-
ing progress has been made on strategies
to use NMR directly to follow folding. In
principle such experiments could allow
the detailed analysis of the structural
ensembles populated at different stages
of the folding reaction, and transform
fundamentally the level of detail in
which we are able to define the folding
process.
Kinetic NMR approaches
The use of NMR to study reactions of
proteins in ‘real time’ started nearly 30
years ago with the objective of studying
enzymatic mechanisms
5,6
. Not long
afterwards efforts were made to study
protein folding and unfolding, and a
variety of experimental strategies were
developed for this purpose (reviewed in
ref. 7). Particular emphasis was placed
on slow reactions to overcome the
intrinsic difficulties in accumulating
NMR spectra of adequate quality in
short periods of time. Unfolding reac-
tions frequently take place over minutes
or hours, and some specific types of
Kinetic studies of protein folding using
NMR spectroscopy
Christopher M. Dobson and Peter J. Hore
Recent progress has advanced our abilities to use NMR spectroscopy to follow — in real time — the structural
and dynamic changes taking place during protein folding.
Fig. 1 Stopped-flow
19
F NMR spectra of the refolding of 6-
19
F-tryptophan labeled Escherichia coli dihy-
drofolate reductase following dilution from 5.5 to 2.75 M urea at 5 °C in the presence of 3.8 mM
NADP
+
. The disappearance of the five resonances of the unfolded state, clustered between -46.0 and
-46.6 p.p.m., and the growth of the more widely dispersed native peaks are clearly seen in this well-
resolved set of spectra. Each spectrum represents the sum of 41 separate rapid dilution experiments.
The kinetics and chemical shifts suggest the formation of an intermediate that is unable to bind
NADP
+
strongly, having a native-like side chain environment in the regions around tryptophans 30, 47
and 133, and little if any native side chain environment around tryptophans 22 and 74. The resonance
labeled 47i is that of Trp 47 in the intermediate. (Taken from ref. 10 with permission).
NMR supplement
nature structural biology ? NMR supplement ? july 1998 505
folding reactions (such as those limited
by the need to isomerize peptide bonds
involving proline residues) can also be
very slow. In order to monitor rapid
reactions, stopped flow procedures
involving rapid mixing within the NMR
sample tube are being developed
7,8
.
Experiments of this type have recently
begun to transform NMR into a general
and powerful technique for studying a
wide range of fundamental events in
folding.
One of the obvious requirements in
these studies is obtaining sufficient reso-
lution to be able to monitor events at the
level of single residues. One extremely
powerful approach has been to use
19
F
NMR to study proteins in which specific
residues (particularly aromatic ones)
have been replaced by fluorinated
analogs. This strategy has been pio-
neered by Frieden and coworkers, and
has provided novel insights into both
unfolding and folding reactions
8–10
. The
ability to use one-dimemnsional (1D)
NMR enables data collection to begin
within 100 ms of mixing, and has
allowed, for example, distinct steps in
the folding of dihydrofolate reductase to
be resolved and characterized through
repetitive collection of spectra during
the folding process (Fig. 1)
10
. A similar
strategy is of course possible using two-
dimemnsional (2D) approaches if the
reactions under investigation are suffi-
ciently slow. Baum and colleagues have
exploited this in an extremely elegant
manner to study the folding and assem-
bly of peptide fragments of collagen
11,12
.
By labeling the peptides with
15
N it has
been possible to record 2D HSQC spec-
tra at intervals of as little as four min-
utes, and to observe the transition
of these peptides from disordered
monomers to structured trimers (Fig. 2).
As well as enabling the mechanism of
this process to be defined, these experi-
ments are providing key information
about the molecular basis of diseases
associated with mutations in the gene
encoding the collagen sequence (J.
Baum, pers. comm.).
An attractive alternative presents itself
for proteins whose folding can be initiat-
ed photochemically. A nice example is
the very recent study by Kaptein and
coworkers of photoactive yellow protein
(PYP), the proposed photosensor of the
motile bacterium Ectothiorhodospira
halophilia (R. Kaptein, pers. comm.).
Light excitation induces the trans-cis iso-
merization of the p-coumaric acid cofac-
tor, which triggers a cycle of structural
changes in PYP yielding an intermediate,
pB, that reverts to the native state pG in
~1 s. Characterization of pB by NMR
revealed that it exhibits extensive struc-
tural and dynamic disorder, in strong
contrast to pG. The conversion of pB to
pG can therefore be considered to be a
folding reaction. It was monitored in
detail by observing the recovery of pG
cross peaks in a series of (
1
H,
15
N) HSQC
spectra recorded at different times after a
laser pulse. Considerable variation in the
build-up rates was found, with more
rapid recovery for the more disorganized
regions of the protein. The major excep-
tion to this was in the neighborhood of
the chromophore, where slow refolding
correlates with high degrees of disorder,
suggesting that retro-isomerization of
the chromophore controls the refolding
of that part of the molecule. Although
less generally applicable than stopped
flow methods, rapid photochemical trig-
gering of refolding (for example, using
nanosecond laser pulses) has the poten-
tial to allow monitoring of very rapid
processes
13,14
.
In our laboratories we have focused on
the development of a variety of comple-
mentary NMR methods aimed at
describing at the atomic level the struc-
tural and dynamic changes taking place
during the folding of a protein from its
denatured state. The ultimate objective is
to map out by experiment the ‘energy
surface’ of the folding reaction
1
. This
requires the ability to monitor the envi-
ronments of individual residues during
folding (for example, whether they are
buried or exposed to solvent) and partic-
ularly to define the inter-residue interac-
tions or ‘contacts’ that develop at
different stages of folding. The latter can
in principle be studied directly if nuclear
Fig. 2 NMR folding profiles of a peptide (top) labeled with
15
N at Gly 24 (circles) and Ala 13 (squares).
O is the one-letter amino acid code for hydroxyproline. The central panel shows the time dependence
of the cross peaks in an (
1
H-
15
N) HSQC spectrum of the peptide as it folds to form a collagen-like triple
helix. The disappearance of the monomer peaks (solid lines) and the appearance of the trimer peaks
(dashed lines) are faster for Gly 24 than for Ala 13. Gly 24 and Ala 13 follow 2
nd
and 1
st
order kinetics
respectively. The data are consistent with a mechanism (bottom) involving intermediates in which the
local conformation of Gly 24 towards the chain end is largely helical while the more central Ala 13 is
still in the unfolded state. (Taken with permission from ref. 12).
NMR supplement
506 nature structural biology ? NMR supplement ? july 1998
Overhauser effects (NOEs) can be
detected between specific nuclei. It will
be necessary of course to interpret these
in terms of structural ensembles, as we
have discussed above, and to obtain
information about the dynamic events
associated with the polypeptide chain as
folding takes place.
We have used the family of c-type
lysozymes and their structural
homologs, the a -lactalbumins, as test
systems for many of these experiments
because the folding of these proteins has
been studied in detail using a wide variety
of other biophysical methods
1,15,16
. In addi-
tion, it is possible to alter the folding kinet-
ics of some members of this family by
factors of ~100 simply by changing the
Ca
2+
concentration in the refolding buffer.
This turns out to be an extremely valuable
factor in devising NMR experiments to
probe different aspects of the folding
process. One approach we have adopted to
extract structural information from 1D
experiments is to exploit photo-CIDNP
17
(photochemically induced dynamic
nuclear polarization). This technique, in
which photo-excitation of a dye molecule
can result in enhanced nuclear polariza-
tion of tryptophan, tyrosine and histidine
residues to which it has access
18
has been
used to probe the accessibility of these
residues in both native and denatured
states
19
. We have found that it can be par-
ticularly powerful in time-resolved exper-
iments (Fig. 3). Because polarization is
induced in only a small number of
residues, the resulting spectra are relative-
ly well resolved. The approach also has a
shorter experimantal dead time than con-
ventional NMR, firstly because the polar-
ization is produced during a ~50 ms light
flash, a somewhat faster process than the
spin-lattice relaxation required to polar-
ize spins transferred into the NMR probe
from a lower field region of the magnet.
Secondly, efficient mixing is only needed
in the small portion of the sample
exposed to the laser flash, from which the
signal is detected
17
.
This rapid mixing approach, coupled
with more conventional 1D experiments
20
has enabled probing of the disordered col-
lapsed state, formed rapidly after the initi-
ation of refolding of these proteins, and
monitoring of the rearrangement process-
es that occur subsequently. We are present-
ly engaged in attempts to increase
significantly the sensitivity of this experi-
ment, and to develop 2D variants. This
task has been substantially aided by the
recognition that it is not necessary to
record sequential spectra to monitor
kinetic events
21
. It turns out that this infor-
mation can be extracted from a single 2D
spectrum recorded while the time-depen-
dent process takes place. If a reaction
occurs during the accumulation of data in
the experiment, it perturbs the line shapes
and intensities of the cross-peaks in the
resulting 2D spectrum. Computer simula-
tion and kinetic model-fitting of these
spectral features gives residue-specific rate
constants for the folding reaction. This
approach has been used already to probe
the cooperativity of the formation of
native-like structure in bovine a -lactalbu-
min during folding using a (
1
H-
15
N)
HSQC experiment
21
(Fig. 4), and to probe
the structure of a folding intermediate
with a non-native proline isomer formed
in the refolding of ribonuclease T1 (J.
Ballach, pers. comm.).
Although such experiments are trans-
forming the possibilities for NMR in study-
ing folding, the detection of NOEs in
collapsed and partially folded states
remains a major challenge. In many cases
the intrinsic problems of dealing with het-
erogeneous systems are compounded by
the existence of extensive exchange broad-
ening of resonances resulting from slow
interconversion of conformers within com-
pact but disordered systems. Nevertheless,
we have been able to show that extensive
NOEs do exist at the earliest detectable
stages of the folding of a -lactalbumin, and
their characteristics indicate that the mole-
cules have native-like compactness
22
. In
order to begin to identify the specific NOEs
within such species we have made use of
the fact mentioned above that Ca
2+
can
profoundly change the folding kinetics of
the protein. The idea is to generate NOEs in
the partially folded state, and then to refold
the protein rapidly to its native state
22
.
Provided this refolding can be done rapidly
compared with the nuclear relaxation rates,
it is possible to transfer the NOEs to the
well resolved spectrum of the native state
for detection. Initial attempts to implement
Fig. 3
1
H photo-CIDNP spectra (B) of the refolding of hen lysozyme after a simultaneous pH jump
from 1.1 to 5.2, and dilution from 10 to 1.4 M urea
17
. The delay times are the intervals between the
end of the 30 ms injection of protein solution into the refolding buffer and the beginnning of the
50 ms light flash that generates photo-CIDNP. Each spectrum is the result of a single injection. The
first spectrum, at 30 ms, differs markedly from that of the denatured state in 10 M urea (A) and
suggests a rapidly formed disordered collapsed state which reorganizes within a second to gener-
ate the native state whose spectrum is shown as (C).
NMR supplement
nature structural biology ? NMR supplement ? july 1998 507
this radio frequency pulse-labeling method
indicate the viability of this approach, sug-
gesting that further development of this
strategy could result in the information
required to characterize the protein folding
process in detail
22
. We are also developing a
closely related experiment in which accessi-
ble aromatic side chains in the partially
folded state are labeled by photo-CIDNP
and then identified in the native state after
rapid refolding.
Concluding remarks
We have outlined briefly in this article
some of the approaches that are being
developed to exploit the potential of
NMR in kinetic studies of protein fold-
ing. These experiments are in the early
stages of development and there are
many opportunities for major advances
in the future, including novel methods of
initiating the folding process, for exam-
ple by using electrochemical or photo-
chemical techniques
13,14
, or temperature
jumps
23
, as well as innovations in NMR
methodology. The information emerging
from these experiments is highly comple-
mentary to that emerging from NMR
studies of equilibrium folding intermedi-
ates
24
, including those that probe the sta-
bilities of interactions by subjecting these
species to progressive denaturation
25
. In
addition to the real-time kinetic experi-
ments described here, there are exciting
NMR approaches based on magnetiza-
tion transfer and line shape analysis that
are providing information about folding
kinetics, including events taking place on
microsecond time scales
26
. In combina-
tion with other experimental approaches
and theoretical advances, NMR is likely
to play an increasingly important role in
the quest to understand the mechanisms
by which proteins fold.
Acknowledgments
We thank J. Balbach, V. Forge, J. A. Jones, N. A. J.
van Nuland and S. L. Winder, for many of the ideas
that have gone into the work from our own
laboratories and that is discussed here. We are
grateful to J. Baum, H. B. Gray, R. Kaptein and J.
Balbach for kindly providing reprints of
unpublished work, and to C. Freiden and J. Baum
for copies of Figs 1 and 2. The Oxford Centre for
Molecular Sciences is supported by BBSRC, EPSRC
and MRC. The research of C.M.D. is also supported
by the Howard Hughes Medical Institute and the
Wellcome Trust.
Christopher M. Dobson is at the Oxford
Centre for Molecular Sciences, New
Chemistry Laboratory and Peter J. Hore is
at the Physical and Theoretical Chemistry
Laboratory, University of Oxford, South
Parks Road, Oxford OX1 3QR, UK.
Correspondence should be addressed to
C.M.D. email: chris.dobson@chem.ox.ac.uk
or P.J.H. email: peter.hore@chem.ox.ac.uk
1. Dobson, C.M., Sali, A. & Karplus, M. Angew. Chem.
Int. Ed. Eng.37, 868–893 (1998).
2. Smith, L.J, Fiebig, K.M., Schwalbe, H. & Dobson, C.M.
Folding & Design 1, 95–106 (1996).
3. Plaxco, K. & Dobson, C.M. Curr. Opin. Struct. Biol. 6,
630–636 (1996).
4. Baldwin, R. L. Curr. Opin. Struct. Biol. 3, 84–91
(1993).
5. Grimaldi, J. J., Baldo, J., McMurray, C. & Sykes, B. D.
J. Amer. Chem. Soc. 94, 7641–7645 (1972).
6. Grimaldi, J. J. & Sykes, B. D. Rev. Sci. Instr. 46,
1201–1205 (1975).
7. van Nuland, N.A.J., Forge, V., Balbach, J. & Dobson,
C.M. Accts. Chem. Res,, in the press.
8. Frieden, C., Hoetzli, S. D. & Ropson, I. J. Prot. Sci. 2,
2007–2014 (1993).
9. Hoetzli, S.D. & Frieden, C. Biochemistry 35,
16843–16851 (1996).
10. Hoetzli, S.D. & Frieden, C. Biochemistry 37, 387–398
(1998).
11. Liu, X., Siegel, D. L., Fan, P., Brodsky, B. & Baum, J.
Biochemistry 35, 4306–4313 (1996).
12. Baum, J. & Brodsky, B. Folding & Design 2, R53–R60
(1997).
13. Pascher, T., Chesick, J. P., Winkler, J. R. & Gray, H. B.
Science 271, 1558–1560 (1996).
14. Telford, J. R., Wittung-Stafshede, P., Gray, H. B. &
Winkler, J. R. Acc. Chem. Res. (1998) in the press.
15. Dobson, C. M., Evans, P. A. & Radford, S. E. Trends
Biochem. Sci. 19, 31–37 (1994).
16. Kuwajima, K. FASEB J. 10, 102–109 (1996).
17. Hore P. J., Winder S. L., Roberts, C. H. & Dobson, C.
M. J. Amer. Chem. Soc. 119, 5049–5050 (1997).
18. Hore, P. J. & Broadhurst, R. W. Prog. NMR Spec. 25,
345–402 (1993)
19. Broadhurst, R. W., Dobson, C. M., Hore, P. J.,
Radford, S. E. & Rees, M. L. Biochemistry 30, 405–412
(1991).
20. Balbach, J. et al. Nature Struct. Biol. 2, 865–870
(1995).
21. Balbach, J. et al. Science 274, 1161–1163 (1996).
22. Balbach, J. et al. Proc. Natl Acad. Sci. USA 94,
7182–7185 (1997).
23. N?lting, B., Golbik, R. & Fersht, A. R. Proc. Natl.
Acad. Sci. USA 92, 10668–10672 (1995).
24. Dyson, H.J. & Wright, P.E. Nature Struct. Biol. 5,
499–503 (1998).
25. Schulman, B., Kim, P. S., Dobson, C. M. & Redfield, C.
Nature Struct. Biol. 4, 630–634 (1997).
26. Huang, G. S. & Oas, T. G. Proc. Natl. Acad. Sci. USA
92, 6878–6882 (1995)
Fig. 4 (
1
H-
15
N) HSQC spectra of bovine a -lactalbumin at 3
o
C during different stages of the folding process. a, Poorly resolved spectrum of the dena-
tured state (A-state) at pH 2.0 recorded before the initiation of refolding. b, Kinetic spectrum accumulated during folding (30 min). c, Well resolved
spectrum of the native (N) state at pH 7.0 recorded after the refolding reaction. The insets show enlargements of the region containing the Val 92
resonance of the N-state. The lower intensity of this resonance in spectrum (b) compared to (c), and the negative features above and below the cen-
tral peak contain information on the local rate of formation of native structure
21
.
a b c