an
o
Chemistry Department,
Lensfield Road, University
of Cambridge, Cambridge
CB2 1EW, UK
other studies.
doi:10.1016/j.jmb.2007.04.039© 2007 Elsevier Ltd. All rights reserved.
*Corresponding author Keywords: EX1; two-state model; three-state model; NMR; protein folding
Introduction
Green fluorescent protein (GFP) from the jellyfish
Aequorea victoria is one of the most important
proteins currently used in biological and medical
research having been extensively engineered for use
as a marker of gene expression and protein
localization, as an indicator of protein–protein† J.-r. H. and T. D. C. contributed
Abbreviations used: GFP, green fl
H/D, hydrogen/deuterium; GdmC
chloride.
E-mail address of the correspondi
sej13@cam.ac.uk
0022-2836/$ - see front matter © 2007 EWe present a study of the denaturation of a truncated, cycle3 variant of
green fluorescent protein (GFP). Chemical denaturation is used to unfold
the protein, with changes in structure being monitored by the green
fluorescence, tyrosine fluorescence and far-UV circular dichroism. The
results show that the denaturation behaviour of GFP is complex compared
to many small proteins: equilibrium is established only very slowly, over the
time course of weeks, suggesting that there are high folding/unfolding
energy barriers. Unfolding kinetics confirm that the rates of unfolding at
low concentrations of denaturant are very low, consistent with the slow
establishment of the equilibrium. In addition, we find that GFP significantly
populates an intermediate state under equilibrium conditions, which is
compact and stable with respect to the unfolded state (mIU=4.6 kcal mol−1
M−1 and ΔGIU=12.5 kcal mol−1).
The global and local stability of GFPwas probed further bymeasuring the
hydrogen/deuterium (H/D) NMR exchange rates of more than 157
assigned amide protons. Analysis at two different values of pH showed
that amide protons within the β-barrel structure exchange at the EX2 limit,
consequently, free energies of exchange could be calculated and compared
to those obtained from the denaturation-curve studies providing further
support for the three-state model and the existence of a stable intermediate
state. Analysis reveals that amide protons in β-strands 7, 8, 9 and 10 have,
on average, higher exchange rates than others in the β-barrel, suggesting
that there is greater flexibility in this region of the protein. Forty or so amide
protons were found which do not undergo significant exchange even after
several months and these are clustered into a core region encompassing
most of the β-strands, at least at one end of the barrel structure. It is likely
that these residues play an important role in stabilizing the structure of the
intermediate state. The intermediate state observed in the chemical
denaturation studies described here, is similar to that observed at pH 4 inand Sophie E. Jacks
othy D. Craggs†, John Christodoulou
n
⁎Jie-rong Huang†, TimStable Intermediate States
the Unfolding of GFPequally to the work.
uorescent protein;
l, guanidinium
ng author:
lsevier Ltd. All rights reserved High Energy Barriers in
J. Mol. Biol. (2007) 370, 356–371interactions and as a biosensor.1 Its widespread
use results from its unique spectroscopic properties,
the 238 residue protein undergoing an autocatalytic
post-translational cyclization and oxidation of the
polypeptide chain around residues Ser65, Tyr66 and
Gly67, to form an extended conjugated π-system,
the chromophore, which emits green fluorescence.2
No cofactors are necessary for either the formation
d.

or the function of the chromophore,3 which is
embedded in the interior of the protein surrounded
by an 11-stranded β-barrel4,5 (Figure 1). GFP is
remarkable for both its structural stability, and high
fluorescence quantum yield, the latter a result of the
fact that in the native state the chromophore is rigid
and shielded from bulk solvent. Upon denaturation,
the GFP chromophore remains chemically intact but
fluorescence is lost with the destruction of tertiary
structure. The green fluorescence is therefore a
sensitive probe of the folding of the protein.6
In all cases, GFP needs to fold efficiently in order
to function in the myriad of biological assays and
experiments in which it is used, and inefficient
folding is known to limit its use in some appli-
cations.1 Despite this, relatively little is known about
the folding of this protein either in vitro or in vivo.
Recent studies by Kuwajima and co-workers have
provided the most detailed information to date.7,8
Their studies have focused on the folding pathway
of GFP from the acid-denatured state and they have
proposed a model in which GFP folds through
several intermediate states. Although, this paper
represents a significant contribution to our under-
standing of the folding of GFP, further studies using
complementary techniques and probes are clearly
ding of the protein under equilibrium conditions.11
These techniques have been used extensively on a
number of proteins and have provided valuable
information ondifferent aspects of the folding energy
landscapes. H/D exchange can inform on partially
structured states (potential high energy intermedi-
ates on a folding pathway),12–14 on global and local
stability,15–18 on residual structure in the denatured
state,19–24 as well as on cooperatively unfolding
regions of proteins.11,14 Although H/D exchange
results have been reported for GFP, the conditions
used did not allow a quantitative analysis of the
results in terms of the global and local stability of the
protein.25
Here, we have applied both optical spectroscopy
and H/D exchange NMR experiments to study the
global and local stability of GFP. Fluorescence, far-
UV CD and NMR measurements are made under
the same conditions at different pH values and
temperatures, enabling not only a comparison
between the two probes, but also the establishment
of the H/D exchange regime (EX1/EX2). At the EX1
limit, where the intrinsic exchange constants are
high in comparison to the closing rates, then the
amide exchange rates are determined by the open-
ing rates, which, for amide groups which only
357Intermediate States in the Unfolding of GFPnecessary in order to provide a more complete
description of the folding pathway of this large,
complex and important protein.
A complete assignment of the NMR resonances
for the backbone (13C, 15N and 1H) of GFP has been
published independently by ourselves and others9,10
thus enabling the use of hydrogen/deuterium (H/D)
exchange techniques to probe the stability and fol-Figure 1. Schematic representations for the structure of GF
LLC), viewed from two opposite sides. The chromophore is sh
to the C terminus.exchange on global unfolding, correspond to the
unfolding rate.15 In contrast, at the EX2 limit, where
the intrinsic exchange constants are small in
comparison with the closing rates, then the
exchange rate observed is determined by the ratio
of the opening and closing rate constants (the
equilibrium between open and closed states) and
the intrinsic exchange rate constant which dependsPuv (PDB code: 1B9C) drawn by PyMol (DeLano Scientific
own in stick mode. Each β-strand is numbered from the N

eas
due
α-h
358 Intermediate States in the Unfolding of GFPupon pH.15 Therefore, under EX2 conditions,
thermodynamic data on the transition between
open (unfolded) or closed (folded) forms of the
protein can be obtained. Establishment of the
exchange regime for GFP enables a full quantitative
analysis of the exchange rates measured, thus
leading to a complete description of the folding of
GFP under equilibrium conditions. In both sets of
experiments, a rigorous quantitative analysis is used
to reveal the presence of, at least, one intermediate
state. The properties of this intermediate state are
discussed. In addition to the equilibrium experi-
ments, the unfolding kinetics of GFP over a wide
range of chemical denaturant concentrations has
Figure 2. The correlation in exchange rate constants m
Residues within the β-sheet structure (filled circles); (b) resi
the barrel and filled triangles for those inside the barrel) or
EX2 limit, the broken line represents the EX1 limit.been measured to establish that there are high-
energy barriers in the unfolding reaction consistent
with the equilibrium results.
Results and Discussion
H/D exchange NMR experiments
The H/D exchange rates of 157 amide protons in
GFP were measured at pH 6.4 and pH 7.4 at 37 °C
over a period of several months by recording
successive 15N, 1H heteronuclear single quantum
coherence (HSQC) spectra. Amide protons were
found to exchange with a very wide range of rates;
Figure 3. (a) The network of hydrogen bonds in the β-bar
antiparallel β-strands represents two hydrogen bonds. A si
exchange rate constants for the amide groups are classed as
month); yellow, ΔGHX N9.0 kcal mol− 1; green, 7.0 kcal mol− 1b
blue, very fast (amide hydrogen exchanged within 20 min); g
spectrum. (b) Amides in regions of random coil (circles) or α
yellow, exchanged within a month; white, overlapped peaks; g
position of very slow exchanging residues. Red balls represent
helices. The left and right Figures are from the same view poin
the β-barrel.the fastest protons having exchanged within the
dead time of the experiment, whereas the slowest
amides did not exchange significantly even after
several months.
Characterising the exchange regime: EX1 or
EX2 mechanism?
In order to interpret the measured amide
exchange rate constants, and to be able to compare
the H/D exchange data with those obtained from
other experiments, it is important to know whether
amide exchange is at the EX1 or EX2 limit. Several
methods can be used to identify EX1 and EX2
ured at pH 6.4 and pH 7.4 for amide protons of GFP: (a)
s in regions of random coil (filled squares for those outside
elices (open diamonds). The continuous line represents themechanisms, including the dependence of kex on pH,
which is a reliable and widely used method.26–28 A
plot of log kex versus pH for a given amide hydrogen
is diagnostic: at the EX1 limit, kex does not change
with pH and is equal to the opening rate constant,
kop. In contrast, at the EX2 limit, kex varies with pH
because it is a function of the intrinsic exchange rate
constants, kint, which depends on pH. Hydrogen
exchange is primarily a base-catalyzed reaction,15 so
kint increases by one order of magnitude per pH unit.
Therefore, at the EX2 limit, log kex would increase
with pH in a linear fashion with a slope of one. One
disadvantage of this method, however, is that it
assumes the stability of the native state does not
vary significantly with pH.29
rel structure. A double line between two residues on two
ngle-line represents one hydrogen bond. The measured
follows. Red, very slow (the half-life is longer than one
ΔGHX b9.0 kcal mol− 1; blue, ΔGHX b7.0 kcal mol− 1; dark
rey, not assigned; white, overlapped peaks in the HSQC
-helices (rectangles). Red, very slow; dark blue, very fast;
rey, not assigned. Three-dimensional representation of the
VS residues in β-strands. Yellow balls represent those in α-
t as used in Figure 1. The central Figure is from the top of

An alternative approach is used here which is an
extension of the method described above, and
involves comparing the kex values of different
protons at just two different pH values.14,30–32 We
measured theH/D exchange rate at two different pH
values , pH7.4 and pH6.4 and by plotting log kexpH 6.4
359Intermediate States in the Unfolding of GFPFigure 3 (legend on previous page)

versus log kexpH 7.4, we can distinguish between the two
limits. At the EX2 limit, we should obtain a straight
line intersecting with the y-axis at one and with a
gradient of one. At the EX1 limit, we should obtain a
straight line with a gradient of one, which would
intersect the y-axis at zero, as the exchange rate
constants are independent of pH under these
conditions.
Figure 2(a) shows the correlation in exchange
rate constants measured at pH 6.4 and pH 7.4 for
GFP amides in the β-sheet structure. It is clear
from this plot that almost all the residues within
the β-sheet exchange at the EX2 limit. The only
exception is Gly104, however, on close inspection
of the crystal structure33 (Protein Data Bank
accession code: 1B9C), it can be seen that its
amide NH group is not involved in hydrogen-
bond formation in the β-sheet. Only the carbonyl
oxygen of Gly104 is H-bonded to the amide
hydrogen of Phe100, while the carbonyl oxygen
of Phe100 is hydrogen-bonded to the amide
hydrogen of Asp103.
The correlation in exchange rate constants mea-
sured at pH 6.4 and pH 7.4 for amides in regions of
coil or α-helix in GFP is shown in Figure 2(b). In
comparison with the β-sheet residues, there are
significantly fewer data points for coil and helical
regions; this is largely because many of these are
exposed on the surface of the protein and undergo
extremely rapid exchange which cannot be mea-
sured. Data are coded to show: (i) residues in
α-helical regions; (ii) residues in coil buried inside
the β-barrel; and (iii) other residues. Many of the
α-helical residues (shown by open diamonds in
Figure 2(b)) exchange at the EX1 limit, although
these are residues at the end of α-helices. In contrast,
the residues in the middle of helices (Thr59, Leu60,
Lys85) exchange extremely slowly and accurate rate
constants for these residues cannot be calculated. In
comparison with the data on amides involved in
β-sheet formation that clearly show EX2 behaviour
(Figure 2(a)), the results for amide protons in regions
of coil or α-helix are somewhat ambiguous. Al-
though, in theory, kinetic information can be
obtained from amides which follow an EX1 me-
chanism,29 we focus here on the amide groupswhich
are known to exchange at the EX2 limit in order to
undertake a thermodynamic analysis. All the follow-
ing discussion is, therefore, based on the β-sheet
amides, which form the barrel structure of GFP.
Calculating ΔGHX from H/D exchange data
Because most of the residues in the 11-stranded
β-barrel structure exchange at the EX2 limit, equa-
tion (3) can be used to calculate a free energy
associated with the exchange process, ΔGHX, for
each residue. The calculated values of ΔGHX of
residues within the β-sheets are between 6–11 kcal
mol−1. These are the limits of ΔGHX, which we can
360accurately calculate under the conditions used. If
ΔGHX is less than 6 kcal mol−1, the half-life would be
less than 3 min, and significantly shorter than thetime it takes to acquire the first spectrum (ca 20 min).
If ΔGHX is greater than 12 kcal mol−1 the half-life is
longer than 36 days and the precision of ΔGHX from
HX experiments is less reliable because of the
tendency of GFP to aggregate at high temperature
and the very high concentrations required in the
NMR experiments over very long periods. Accord-
ingly, we define two classes of exchange: (1) very
fast (VF), where complete exchange occurs before
the first spectrum is acquired; and (2) very slow
(VS), where the intensity of the cross-peak is still
greater than 50% its original value after one month
of exchange, i.e. the half-life is longer than one
month. For all the other amides measured, the
exchange rate constants are converted to values of
ΔGHX which are then categorized into three classes:
(i) ΔGHX less than 7.0 kcal mol−1; (ii) ΔGHX between
7.0 and 9.0 kcal mol−1; and (iii) ΔGHX greater than
9.0 kcal mol−1. In addition to VS and VF, we
therefore have three additional classes of exchange,
i.e. a total of five classes. These are mapped onto the
structure of GFP (Figure 3(a)) and color-coded
according to class. A wide range of values of ΔGHX
are observed as has been reported for other proteins;
hydrogen exchange is caused by both local fluctua-
tions and global/sub-global unfolding events in the
absence of denaturant.12 All the very slowly
exchanging amide hydrogen atoms are likely to
exchange via a mechanism which is dominated by
the global unfolding event.
There are about 40 peaks in the HSQC spectra,
which exchange in the VS category (Figure 4). The
ΔGHX of these very slowly exchanging cross-peaks
can be estimated to be greater than 12 kcal mol−1,
and the largest ΔGHX can be regarded as corre-
sponding to global unfolding.34 The exchange rate
constants of these very slowly exchanging amide
hydrogen atoms can be estimated to be in the order
of 10−8∼10−9 s−1 (see Figure 5). In these cases, the
ΔGHX for global unfolding can be estimated to be
between 14 kcal mol−1 and 15.5 kcal mol−1.
AnH/D exchange study of GFP has been reported
by Holak and co-workers.25 In their NMR study,
measurements of H/D exchange were used to
provide information on the conformational flexibil-
ity in the protein on the micro-to millisecond time-
scale and these were compared with 15N-relaxation
measurements which report on pico-to nanosecond
time-scale fluctuations. Whereas, the latter showed
that the β-barrel of GFP is rigid on short time-scales,
the H/D exchange showed that there is increased
flexibility in β-strands 3 and 7–10 over longer time-
scales. In our study, we repeated the H/D exchange
measurements on GFP under controlled conditions,
which allow us to extend the analysis of the
exchange rates and enable us to calculate quantita-
tive values of stability in terms of ΔGHX.
Both our study and that of Holak and co-
workers,25 show that the region of β-barrel encom-
passing strands β7 to β10 has, on average, higher
Intermediate States in the Unfolding of GFPexchange rates than the rest of the β-barrel structure.
However, even within these four strands there are
some amides, which exchange only very slowly

ex
Intermediate States in the Unfolding of GFPFigure 4. The HSQC spectrum after one month of H/D(Figure 3(a)). It is, therefore, not completely clear
whether the antiparallel four-stranded β-sheet
formed by β-strands β7 to β10 is a cooperatively
unfolding unit. In addition to the region between β7
and β10, we find that there is no very slowly
exchanging amide between β-strands 2 and 3,
suggesting that these strands might also be able to
move apart from each other without losing their
interactions with the rest of the β-barrel.
A three-dimensional representation of GFP
labelled with the most slowly exchanging residues
Figure 5. Exchange data for the very slowly exchan-
ging amide group of Tyr106. The continuous curves
represent the data expected for exchange rate constants
of 10− 6–10− 9 s− 1. This allows an estimation of the exchange
rate constant for Tyr106 of between 10− 8–10− 9 s− 1.
peaks of Leu42, Leu44, and Cys48; the asterisk (*) is the overl
spectrum of GFP before H/D exchange.change. Hatch mark (#) are used to indicate the cluster of
361is shown in Figure 3(c). In general, these residues are
clustered together primarily on one face of the β-
barrel; however, at one end of the barrel they do
form a ring-like structure. It is likely that these
residues exchange only very slowly because they
remain involved in hydrogen-bonded secondary
structure in the intermediate state (see the next
section).
Chemical denaturation and fluorescence
measurements
Guanidinium chloride (GdmCl) was used to in-
duce the unfolding of GFP and denaturation curves
were monitored by green fluorescence, a sensitive
probe of the native state,6–8 over a wide range of
equilibration times. In contrast to chemical denatura-
tion studies on small monomeric proteins, which
reach equilibrium rapidly, we observe that GFP
reaches equilibrium only very slowly (Figure 6).
Unusual though this may seem, it is not unprece-
dented and has been reported in a number of studies
on the unfolding of fluorescent proteins.35,36 In these
two studies, the stability of oligomeric fluorescent
proteins (FPs) such as DsRed were compared with
EGFP and stability changes over six to nine days
reported.
Despite the fact that the system is not fully
equilibrated, it is possible to fit each denaturation
curve to various models of folding and obtain
apparent thermodynamic parameters. It should be
apped peak of H169 and I188. The inset shows the HSQC

362stressed that the analysis does not assume that the
system is at equilibrium; we are merely fitting the
curves to equations that measure the relative
Figure 6. Chemical denaturation curves monitored by
green fluorescence. (a) Green fluorescence was measured
as a function of GdmCl concentration at pH 7.4 at 37 °C
over a wide range of equilibration times: 14 h (filled
squares), 24 h (open squares), 48 h (filled triangles), one
week (open triangles), two weeks (filled diamonds), three
weeks (open diamonds), four weeks (filled circles), five
weeks (open circles), sixweeks (crosses), sevenweeks (plus
signs), and eight weeks (asterisks). (b) Green fluorescence
wasmeasured as a function of GdmCl concentration at pH
6.4 at 37 °C over a wide range of equilibration times. The
symbols are the same as for (a) except the first three time
points are: 24 h (filled squares), 48 h (open squares), and
96 h (filled triangles). (c) Green fluorescence wasmeasured
as a function of GdmCl concentration at pH 6.0 at 25 °C
over a wide range of equilibration times: 3 h (filled
squares), 12 h (filled triangles), 24 h (reverse filled
triangles), 48 h (filled diamonds), five days (filled circles),
13 days (open squares), and 44 days (open triangles).populations of native, intermediate and denatured
states.
The denaturation curves obtained at pH 7.4, 37 °C
(Figure 6(a)) were first analyzed with a two-state
model and the dataset at each time point fit to
equation (4) to generate values for [D]50%
app andmNU as
a function of denaturation time. The results are
shown in Figure 7(a) and (b). In this case, [D]50%
app
decreases exponentially with equilibration time;
however, the mNU values vary significantly with
time. The entire dataset was also fit globally allowing
[D]50%
app to vary with equilibration time, but sharing
the same mNU for all time points. The results for
[D]50%
app did not differ significantly from the original
individual fitting (Figure 7(c)). An estimate of the
apparent free energy of unfolding,ΔGNU
app, after eight
to nine weeks of equilibration, was obtained by
fitting the data in Figure 7(c) to a single-exponential
decay to obtain a value [D]50%
app at eight to nineweeks,
and then using the relationship ΔGNU
app=mNU[D]50%
app
to give a value of 4.0(±0.3) kcal mol−1. This value is
considerably smaller than the estimated free energies
for global unfolding obtained from the H/D
exchange data.
Although the fluorescence data at pH 7.4 at 37 °C,
and that at pH 6.0 measured at 25 °C (Figure 6(c)), fit
reasonably well to a two-state model, there is clear
evidence from the data obtained at pH 6.4 and 37 °C
(Figure 6(b)), where two transitions can be seen, that
the unfolding of GFP is, at least, a three-state process.
In fact, even at pH 7.4, a slight discrepancy in the
two-state fit to the data can be observed (Figure 7(d)),
and the data are more consistent with a three-state
model (Figure 7(e)). The presence of an intermediate
state can also explain why the apparent free energies
from the fluorescence and NMR experiments do not
agree.
Soulages has calculated the potential error in
estimates of values for ΔGNU and mNU due to the
presence of undetected intermediates from fitting
data to a two-state model.37 Two cases were consi-
dered: (1) native and intermediate states are indis-
tinguishable by the observed unfolding property, for
example, fluorescence; and (2) intermediate and
unfolded states are indistinguishable. Each of these
cases underestimateΔGNU by different amounts and,
therefore, the ΔGNU of GFP calculated as described
above from the fluorescence data using the two-state
model is likely tounderestimate the real value.Mayne
and Englander38 have also shown from their studies
on cytochrome c that the presence of partially folded
intermediates populated within the transition region
of denaturation curves can affect results quite
dramatically, and the denaturation curves were,
therefore, re-analyzed with a three-state model,
allowingΔGNI
app andΔGIU
app to varywith equilibration
time, but sharing mNI and mIU between all datasets
obtained under the same experimental conditions.
Plots of [D]NI50%
app and [D]IU50%
app against equilibration
time are shown in Figure 7(f). An estimate of ΔGNI
app
app
Intermediate States in the Unfolding of GFPandΔGIU after eight to nineweeks equilibration time
was obtained by fitting these curves to a single-expo-
nential process, and using the equations ΔGNIapp=mNI

Intermediate States in the Unfolding of GFP[D]NI50%
app and ΔGIU
app=mIU[D]IU50%
app to yield values of
3.7(±0.3) kcal mol−1 and 12.5(±0.4) kcal mol−1,
respectively. An estimate of the overall stability of
the native state of GFP with respect to the denatured
state is, therefore, approximately 16 kcalmol−1. These
values are much more consistent with the exchange
Figure 7. Analysis of the chemical denaturation curves sh
[D]50%
app was calculated from the best fit of the denaturation curv
line is the best fit of the data to an equation describing a single-
mNU was calculated from the best fit of the denaturation curves
equilibration time. The [D]50%
app was calculated from the global
and 37 °C to a two-state model. (d) Single denaturation curv
37 °C after four weeks equilibration time showing the raw
(continuous line). (e) Simulated denaturation curves from the
(open circles). The parameters for generating these curves ar
mol− 1. Three-state: mNI=2 kcal mol− 1 M− 1; mIU=4 kcal mol−1
shows the small deviation between the experimental data obse
versus equilibration time. [D]NI50%
app (filled squares) and [D]IU50%
app
pH 7.4 and 37 °C to a three-state model sharing mNI and m
exponential fits to these two datasets.363free energies calculated from the H/D exchange data.
The apparent thermodynamic parameters obtained
from these fits are listed in Table 1. This analysis was
also undertaken for the data obtained at pH 6.0 and
25 °C and we observed very similar results (data not
shown).
own in Figure 6. (a) [D]50%
app versus equilibration time. The
es at pH 7.4 and 37 °C to a two-state model. The continuous
exponential process. (b) mNU versus equilibration time. The
at pH 7.4 and 37 °C to a two-state model. (c) [D]50%
app versus
fit (sharing m-values) of the denaturation curves at pH 7.4
e for GFP monitored by green fluorescence at pH 7.4 and
data (open circles) and the best fit to a two-state model
two-state model (filled squares) and a three-state model
e: two-state: mNU=4.5 kcal mol− 1 M− 1; ΔGNU=11.3 kcal
M− 1; ΔGNI=6 kcal mol− 1; ΔGIU=9 kcal mol− 1. This Figure
rved and the fit when a two-state model is used. (f) [D]50%
app
(filled triangles) are from the global fit of data acquired at
IU between all datasets. The continuous lines are single-

Cl
[D
1
1
U (k
4
4
us e
icThe denaturation curves measured at pH 6.4 and
37 °C (Figure 6(b)) clearly will not fit a two-state
mechanism, and fitting to a three-state model was
undertaken. In this case, however, even global fits to
a three-state model proved difficult (data not
shown). One reason for this is that another
intermediate state may be populated under these
conditions.
From fitting of the fluorescence data to both two
and three-state models, and from a comparison of
the free energies of unfolding obtained from the
H/D exchange and chemical denaturation data, it is
apparent that the equilibrium unfolding of GFP is
three-state. In order to obtain further information on
the unfolding transitions and the intermediate state,
we measured the unfolding curves using two other
optical probes, far-UV circular dichroism (CD) and
tyrosine fluorescence (Figure 8). Due to limitations
with protein sample, only nine time points were
taken with the tyrosine fluorescence, and three with
far-UV CD, but both show similar behaviour to that
observed with green fluorescence, i.e. a slow equi-
libration of the protein in GdmCl. A comparison of
the normalized data measured using the three
different optical probes under the same experimen-
tal conditions and at the same equilibration times
are shown in Figure 8(c)–(e). It is clear from these
data that the transition monitored by green fluores-
cence is consistently at slightly lower concentrations
of denaturant compared with the transitions mon-
itored by tyrosine fluorescence and far-UV CD,
consistent with a three-state model in which an
Table 1. Apparent thermodynamic parameters from Gdm
mNU (kcal mol−1 M−1)
2-stateb No constraint No trend
Shared m-value 2.38±0.04
mNI (kcal mol−1 M−1) mI
3-stateb Shared m-values 1.89±0.06
3-statec Shared m-values 2.60±0.07
a These values were obtained from fitting the plots of [D]50%
app vers
relationship: ΔG=m[D]50% to obtain estimates of the thermodynam
b At 37 °C and pH 7.4.
c At 25 °C and pH 6.0.
364intermediate state is populated, which has lower
green fluorescence than the native state but which
still has considerable structure (as indicated by
tyrosine fluorescence, which probes tertiary struc-
ture throughout the protein, and far-UV CD, which
probes secondary structure). However, the two
transitions (native to intermediate and intermediate
to denatured) occur at similar denaturant concen-
trations and are overlapping.We did not try to fit the
tyrosine or far-UV CD data to a three-state model as
the signal-to-noise was considerable with these
probes compared with the green fluorescence.
In order to ensure that the slow decrease in green
fluorescence signal, tyrosine fluorescence and far-
UV CD signal observed over time was not due to
either the chemical modification of the chromophore
or degradation of the protein sample over the longincubation periods of the denaturation experiments,
absorbance and fluorescence spectra and SDS–
PAGE were recorded for samples of GFP that had
been incubated for extended periods of time (up to
two years at room temperature in either 3 or 6 M
GdmCl in PBS buffer, with 0.1 mM TCEP). Neither
experiment showed any evidence of degradation of
the protein or chromophore (data not shown), thus
providing strong evidence that the transitions we
observe are due to the slow unfolding of the protein.
In addition, the reversibility of GFP unfolding in
GdmCl and refolding into low concentrations of
denaturant was measured by NMR and fluores-
cence spectroscopy. A sample of GFP was incubated
in high concentrations of GdmCl overnight to
ensure complete unfolding and then refolded into
0.1 M GdmCl. A 1-D 1H spectrum of the refolded
GFP was acquired and compared with a spectrum of
a native sample of GFP under identical buffer
conditions and protein concentrations. The results,
shown in the Supplementary Data, establish that
GFP can be reversible refolded with N95% efficiency
under these conditions. An emission spectrum of
this refolded GFP sample confirmed this and
showed that more than 90% of the GFP refolds
under these conditions to a state which has the same
spectral properties as the native protein (see
Supplementary Data).
Nature of the intermediate state – m-value
calculations
denaturation curves
]50% (M)
a
ΔGNU (kcal mol−1)a
.64±0.06 3.54±0.40
.68±0.08 4.00±0.26
cal mol−1 M−1) ΔGNI (kcal mol−1)a ΔGIU (kcal mol−1)a
.56±0.08 3.7±0.3 12.5±0.4
.24±0.31 6.0±0.2 10.5±0.4
quilibration time to a single-exponential decay and then using the
unfolding parameters at equilibrium.
Intermediate States in the Unfolding of GFPIn order to get a better understanding of the
intermediate state, in particular to get some estimate
of how structured this state is compared to the
native state, we calculated theoretical m-values for
several possible intermediate structures. This was
achieved by using SASA calculations for both folded
and unfolded regions of GFP,39–41 combined with
empirical data linking changes in SASA with m-
values42 as described in Materials and Methods. We
were guided in the generation of possible inter-
mediate structures by the H/D exchange results,
assuming that regions which showed increased
levels of exchange compared to average values, are
less stable and are more likely to be unfolded in an
intermediate state. We generated potential models
of intermediates in which the following elements of
secondary structure are unfolded: I7–10 (β7 to β10

Intermediate States in the Unfolding of GFPare unfolded), I7–9 (β7 to β9 are unfolded) or I7–8
(β7 to β8 are unfolded). The results are shown in
Table 2. The theoretical mNU value for complete
unfolding of GFP in GdmCl predicted in this way is
6.32 kcal mol−1 M−1, very similar to the experi-
mental value measured (6.45 kcal mol−1 M−1; Table
1), thus giving validity to the method. The experi-
mental value for mIU, 4.56 kcal mol−1 M−1, is most
similar to the theoretical mIU value calculated for the
Figure 8. (a) The denaturation curves of tyrosine fluorescen
48 h (reverse filled triangles),one week (open squares), two
weeks (open triangles), five weeks (open diamonds), six week
The 48 h and seven week data were fitted with two-state mod
by far-UVCD after equilibration times of twoweeks (filled squ
(c) Comparison of the denaturation curves after two weeks
(reverse triangles), tyrosine fluorescence (squares) and far-U
denaturation curves after three weeks equilibration with differ
fluorescence (squares) and far-UV circular dichroism (triangl
weeks equilibration with different probes. Green fluorescence
UV circular dichroism (triangles). As GFP is monomeric under
concentration dependence to the unfolding curves. Therefore
can be compared directly despite the fact that data are acquir365I7–8 intermediate (4.69 kcal mol
−1 M−1), whereas the
experimental value formNI, 1.89 kcal mol−1 M−1, lies
between the theoretical mNI values calculated for I7–8
and I7–9 (1.63 kcal mol
−1 M−1 and 2.22 kcal mol−1
M−1, respectively). Although these calculations are
only estimates, and do not tell us which elements of
secondary/tertiary structure are unfolded in the
intermediate state, they do provide a good guide for
the extent of structure. The results suggest that at
ce with different equilibration times. 24 h (filled triangles),
weeks (filled diamonds), three weeks (filled circles), four
s (reverse open triangles) and seven weeks (open circles).
el (continuous lines). (b) The denaturation curves followed
ares), three weeks (open squares), and four weeks (crosses).
equilibration with different probes. Green fluorescence
V circular dichroism (triangles). (d) Comparison of the
ent probes. Green fluorescence (reverse triangles), tyrosine
es). (e) Comparison of the denaturation curves after four
(reverse triangles), tyrosine fluorescence (squares) and far-
all the experimental conditions studied, there is no protein
, the green and tyrosine fluorescence and far-UV CD data
ed at different protein concentrations.

GFP which lead to a rather slow equilibration of the
system comparedwith that observed for many small
proteins. To investigate this further, the unfolding
kinetics of GFP as a function of denaturant
concentration were measured under identical con-
lculation of theoretical m-values
ΔSASANU (Å
2) mNU (kcal mol
−1 M−1)
22,582 6.32
ol−1 M−1) ΔSASAIU (Å
2) mIU (kcal mol−1 M−1)
66 13,072 3.66
22 14,652 4.10
63 16,747 4.69
y. I represents the intermediate state. Three possible intermediates are
A between N and U, i.e. (the SASA of U) – (the SASA of N). The same
alculated using equation (6).
Intermediate States in the Unfolding of GFPleast two or perhaps three β-strands are unfolded
and coil in the intermediate state consistent with the
H/D exchange results which show that β-strands
7–10 have increased flexibility. In addition, Helms
et al.43 used a 1 ns molecular dynamic simulation of
GFP to show that the β-barrel is disrupted only
between β7 and β8 strands over these timescales.
The results suggest that not all four strands are
unstructured in the intermediate, and it is therefore
possible that only two to three strands are com-
pletely displaced, or that all four of these strands are
partially displaced. There is some evidence for this
from a careful inspection of the H/D exchange data
which show that there are very slowly exchanging
amide protons in these strands at one end of the β-
barrel (Figure 3(c)).
It is interesting to compare the intermediate we
observe during the GdmCl unfolding of GFP with
that populated at pH 4.8 This acid-induced inter-
mediate has been studied using a range of biophy-
sical techniques including small-angle X-ray scat-
tering measurements and fluorescence spectroscopy
by the Kuwajima group.8 They conclude that the
low pH intermediate is a molten-globule like state in
which the hydrophobic core around Trp57 is
organized but with much reduced levels of quench-
ing of this tryptophan by the GFP chromophore, the
chromophore itself showing little fluorescence. The
SAXS data indicated a compact and globular
structure more expanded than the native state.
Given the current level of structural resolution we
have for the GdmCl-induced intermediate state, the
results suggest that the two stable intermediates
Table 2. Change in solvent accessible surface area and ca
U (Å2) N (Å2)
32,346 9764
I (Å2) ΔSASANI (Å
2) mNI (kcal m
I7–10 19,274 9510 2.
I7–9 17,694 7930 2.
I7–8 15,600 5835 1.
U and N represent the fully unfolded and native states, respectivel
considered here, I7–10, I7–9, and I7–8. ΔSASANU is the change in SAS
rule applies to N and I, I and U. All m-values in this Table were c
366populated under different experimental conditions
may have similar structures: a local unfolding
around β-strands 7–10 would result in an increase
in the overall dimensions of the protein, loss of the
green fluorescence as the chromophore is exposed to
solvent, but with Trp57 remaining largely buried
and in a hydrophobic environment. It should be
noted, however, that higher resolution structural
data on the GdmCl-induced intermediate state is
needed to confirm this idea.
Kinetics studies on the unfolding of GFP
The results from the chemical denaturation stu-
dies described above, suggest that there are high-
energy barriers to the unfolding/folding reactions ofFigure 9. Unfolding kinetics of GFP. (a) Unfolding rate
constants measured at pH 7.4 and 37 °C as a function of
denaturant concentration, measured using green fluores-
cence. The continuous line is the best fit of the data to a
straight line. The half-life for the unfolding of GFP at low
concentrations of denaturant (2 M GdmCl) is estimated to
be approximately 30 days. (b) Unfolding rate constants
measured at pH 6.0 and 25 °C as a function of denaturant
concentration, using different probes and instrumentation.
Values calculated from the GdmCl denaturation curves
shown in Figure 6 by plotting green fluorescence as a
function of time for a particular concentration of dena-
turant are shown in red; kinetic unfolding experiments
using manual mixing and a fluorimeter to measure green
fluorescence are shown in blue and green; unfolding rate
constants measured using a stopped-flow apparatus and
green fluorescence are shown in purple; and unfolding
rate constants measured using manual mixing and a
fluorimeter but using tyrosine fluorescence to probe
folding are shown in yellow. The continuous line shows
the best fit of all the data to a second-order polynomial.

e.g. 5 M GdmCl, the rate of unfolding is reasonably
fast with a half-life of about 30–40 s. However, as the
concentration of denaturant decreases the rate of
unfolding decreases rapidly, the slope of the plot
being 3.6 M−1, considerably steeper than for many
small proteins. Extrapolating the unfolding rate
constant to low concentrations of denaturant, e.g.
1–2 M, which is approximately the midpoint of
denaturation curves after several weeks equilibra-
tion, results in unfolding half lives on the order of
weeks to months. These results are, therefore,
consistent with the chemical denaturation curves,
which reach equilibrium only very slowly.
To investigate the unfolding kinetics further,
experiments were also undertaken at pH 6.0 and
25 °C (Figure 9(b)), using a combination of different
probes and methods. Rate constants obtained using
green fluorescence to probe foldingwere in complete
agreement with those calculated from measure-
ments of tyrosine fluorescence (Figure 9(b)). As
there are ten tyrosine residues in GFP, which are
located in different regions of the structure, tyrosine
fluorescence is a good probe of the tertiary structure
of the protein. Measurements were made over an
extended range of GdmCl concentrations by em-
ploying a variety of methods. At high concentrations
of denaturant, rates were measured using a stopped-
flow apparatus, whilst those at lower concentrations
of denaturant were measured using manual mixing
and a fluorimeter. Finally, rate constants were
calculated at very low concentrations of denaturant
by analysis of the data in Figure 6. The rate constants
calculated by these different methods and probes
were remarkably consistent giving rise to a smooth
curve (Figure 9(b)). In contrast to the data acquired at
pH 7.4 and 37 °C that fit well to a linear equation,
data at pH 6.0 and 25 °C showed significant
curvature and was fit to a second-order polynomial
equation. This apparent difference in behaviour at
25 °C and 37 °C could simply arise from the fact that
data at 25 °C was acquired over a much larger range
of denaturant concentrations. Curvature in the
unfolding limbs of the so-called chevron plots has
now been observed for many proteins and has been
attributed to movement of the transition state
(Hammond behaviour)44 or changes in the rate-
limiting step due to the presence of a high-energy
intermediate.45 As we have strong evidence from the
denaturation curves and H/D exchange NMR data,
that there is a stable intermediate state present
during the unfolding of GFP, we propose that the
curvature observed in the kinetic plots is due to the
presence of this intermediate.
Conclusionditions to the fluorescence and NMR experiments.
The results obtained at pH 7.4 and 37 °C are shown
in Figure 9(a). At high concentrations of denaturant,
Intermediate States in the Unfolding of GFPBy combining measurements of the denaturant-
induced unfolding of GFP monitored by optical
spectroscopies with measurements of the H/Dexchange rates of 157 (nearly two-thirds) of GFPs
amide protons, we have three strong pieces of
evidence to suggest that there is a stable intermedi-
ate state in the unfolding of GFP populated under
equilibrium conditions. This includes the fits of the
green fluorescence chemical denaturation data
measured under three different experimental condi-
tions to two and three-state models, a comparison of
the free energies of unfolding obtained from H/D
exchange experiments and chemical denaturation
studies, and a comparison of the chemical denatura-
tion of GFP with different optical probes.
The intermediate state we observe is compact and
stable with respect to the denatured state with a mIU
value of 4.6 kcal mol−1 M−1 and ΔGIU value of 12.5
kcal mol−1. It is interesting to note that these values
are comparable to the stabilities and m-values
measured for the native state of many small mono-
meric proteins.46 The intermediate retains consider-
able secondary and tertiary structure; however, it has
a reducedgreen fluorescence signal. Froman analysis
of the H/D exchange results and the equilibrium
unfolding m-values, we get some information on the
nature of the intermediate state. From this, we can
propose a possible structure in which some of the β-
strands in the region between β7–β10 have been
displaced and in which there is access of solvent to
the green chromophore. It is interesting to note that it
has been suggested that chromophore formation
takes place in a partially structured intermediate
state, which may be similar to that we observe in our
equilibrium experiments reported here.
In addition, we have established that this large
and structurally complex protein is extremely slow
to unfold at low concentrations of denaturant. This
gives rise to a somewhat unusual behaviour where
the chemical denaturation unfolding curves change
with equilibration time.
Materials and Methods
Protein expression and purification
The gene encoding GFPuv (Clontech) was cloned into a
modified pRSET vector (Invitrogen) without a hexahisti-
dine tag. A stop codonwas introduced at residue 230 using
polymerase chain reaction (PCR) to produce a truncated
form of GFPuv.10 The resulting plasmid (trGFPuv) was
fully sequenced. Thewild-typeGFP referred to throughout
this paper is a pseudo-wild-type GFP corresponding to the
trGFPuv construct described above.
Single colonies of transformed Escherichia coli cells (C41)
harboring trGFPuv were picked from TYE ampicillin plate
and were used to inoculate 5 ml of 2×TY media containing
0.1 mg/ml of ampicillin and were grown overnight in a
shaker at 37 °C. This overnight culture was used to
inoculate 1 l 2×TY media containing 0.1 mg/ml of
ampicillin. The growth cultures were left 37 °C on a
shaker until the cell density at A600 nm was 0.7–0.8. The
temperature was reduced to 25 °C and the cells were
induced with 1 ml of 1 M IPTG and left shaking overnight
367at 25 °C. The fluorescent green cells were harvested by
centrifugation (SLC 4000 Sorvall rotor, 15 min at 8000 rpm
at 4 °C) and resuspended into 50 mM Tris (pH 8.0). The

cells were lysed by sonication on ice for 4 min (15 s pulse
on and 45 s pulse off) power cycles at power level 8 with a
9 mm probe using a sonicator (Misonix Inc). The lysate
was centrifuged at 18,000 rpm (SS34 Sorvall rotor) at 4 °C
for 45 min. The supernatant was then pooled and loaded
onto a Q-Sepharose (Pharmacia) column (1.5 cm by 10 cm)
at 2 ml min−1 until the green protein was visibly bound to
the top of the column. After six column volume washes of
50 mM Tris (pH 8.0), trGFPuv was eluted with a gradient
using 50mMTris (pH 8.0), 0.5MNaCl. The green fractions
were pooled and concentrated (Vivaspins 20; Vivascience).
The concentrated protein (b10 ml) was gel filtrated on a
Superdex G75 column (Amersham Biosciences), which
was pre-equilibrated with phosphate-buffered saline
(PBS) containing 137 mM NaCl, 2.7 mM KCl, 10 mM
phosphate (pH 7.4), and a flow rate of 3 ml min−1. The
purity of the GFP eluted was confirmed by SDS–PAGE.
For NMR studies, a uniformly 15N-labeled sample was
prepared using the same procedure as described above
except the 1 l 2×TYovernight growth media were replaced
by minimal media,47 in which the only source of nitrogen
was 15NH4Cl.
Reagents
Ultra pure guanidinium chloride (GdmCl) was pur-
chased from ICN Biomedicals, Inc. 2H2O (99.96%) and
15NH4Cl were from Cambridge Isotope Laboratories, Inc.
All other chemicals were of analytical grade and pur-
chased form Sigma, BDH, or Melford Laboratories.
Millipore-filtered, double-deionised water was used
throughout.
NMR acquisition and H/D exchange experiments
All NMR spectra were acquired on a Bruker AVANCE
700 MHz spectrometer equipped with 5 mm 1H, 13C, and
15N triple-resonance cryogenic probe heads. Phase-sensi-
tive HSQC) with decoupling during acquisition and 3-9-19
pulse sequence water suppression with gradients48–50 was
used to record the 2-D 1H-15N spectra. All HSQC spectra
were acquired with 1024 (t2) and 256 (t1) complex points.
All spectra were acquired at 37 °C.
H/D exchange experiments were initiated by adding
2H2O (99.96%) to lyophilized protein, which had been
prepared at the required pH and buffer conditions. The
concentration of GFP was about 400 μM (samples tended
to aggregate at concentrations above 600 μM), and the
samples were stored at 37 °C between measurements.
Analysis of H/D exchange data and the EX1/EX2 limit
Exchangeable amide hydrogen atoms (NH) which are
involved in hydrogen-bonded structure can exchangewith
solvent hydrogen atoms only when they are transiently
exposed to solvent in some kind of closed to open reac-
tion.15,16 The Linderstrϕm-LangModel has been applied to
structurally protected hydrogen atoms in H/D exchange
experiments51,52:
NHclosedW
kop
kcl
NHopent
kint
NDopen ð1Þ
where kop and kcl are the opening (unfolding) and closing
368(folding) rate constants, respectively. The intrinsic rate
constant for exchange, kint, depends on the residue type
and various conditions (pH, temperature, neighboringamino acids, and isotope effects) and can be estimated on
the basis of model compound data.53 The exchange rate
constant, kex, can be determined from Scheme (1):
kex ¼
kop  kint
kop þ kcl þ kint
ð2Þ
Two limits of equation (2), so-called EX1 and EX2 limits,
have been derived and described elsewhere.15,16 At the
EX2 limit, ΔGHX can be calculated from the measured
exchange rate constants using the following equation:
DGHX ¼ RT ln Kop ¼ RT ln
kint
kex
ð3Þ
where Kop is the equilibrium opening constant and is equal
to kop/kcl.
When calculating the intrinsic exchange rates of
residues from model compound data,53 we used pD
values of 7.8 and 6.8 (0.4 pH unit above the measured pH
in 2H2O).
Fluorescence measurements
Fluorescence measurements were taken with a SLM-
Aminco Bowman series 2 luminescence spectrometer
using a 1 cm pathlength cuvette. The excitation wave-
length was 395 nm with a band pass of 4 nm for both
excitation and emission. The largest difference in fluores-
cence between the native and unfolded states was
observed at 507 nm, and emission at this wavelength
was used in the subsequent analysis. For tyrosine
fluorescence, the excitation wavelength was 276 nm and
emission wavelength was 307 nm. The volume of each
equilibrium sample was 900 μl with a final concentration
of 1 μM GFP in PBS, 0.1 mM TCEP, and the concentration
of GdmCl varied, typically from 0 to 3–4 M. Measure-
ments were made at pH 6.4 and pH7.4, and the pH value
of the PBS buffer was adjusted by titrating with HCl before
dispensing. Prepared samples were stored in a 37 °C
incubator before fluorescence measurements were made.
For measurements at 25 °C, final conditions were
1 μM GFP, 50 mM Mes( pH 6.0), 0.1 mM TCEP with a
range of GdmCl concentrations from 0–4 M.
Far-UV circular dichroism measuremens
Far-UV CD spectra were recorded using a Jasco J-720
spectropolarimeter with an emission band bass of 2 nm.
Scans were taken between 210 nm and 250 nm at a scan
rate of 1 nm s−1. The largest difference between native and
unfolded state of GFP was observed at 222 nm, which was
used to monitor unfolding. The protein sample was 10 μM
with various concentrations of GdmCl (0 to 4M). Prepared
samples were stored and measured at 37 °C. A 0.1 cm cell
was used.
Two-state and multi-state model analysis
Equations for fitting fluorescence unfolding data to a
two-state model have been derived and described in
detail54:
exp mð½D½D50%Þ


Intermediate States in the Unfolding of GFPF ¼ FN  FN  FUð Þ
RT
1 þ exp mð½D½D50%ÞRT


ð4Þ

where F is the observed intensity of optical property (e.g.
fluorescence) and FN and FU are values of the fluores-
cence intensities of the native and unfolded forms of the
protein, R is the gas constant, T is the absolute
temperature, m is a constant related to the average
fractional change in the degree of exposure of residues on
unfolding, and [D] is the concentration of denaturant.
[D]50% is the midpoint of the unfolding transition and
concentration of denaturant at which 50% of the sample is
unfolded and 50% folded. In practice, the fluorescence
data have sloping baselines. Equation (4) can be modified
to take this into account by substituting, FN=αN+ βN[D],
and FU=αU+ βU[D], where βN and βU are the slopes of
the native and denatured baselines, respectively, and αN
and αU are the intensity of optical properties of the native
state and unfolded state, respectively.55 The resulting
equation is that used to fit the data to a two-state model.
For more complex models, including a three-state
model, a single equation as a function of the concentration
of denaturant can be derived.56 However, using commer-
cially available software, a combination of the equations
described below was used for data analysis. For example,
the following equations describing a three-state model
were used for data analysis within the software package
GraphPad Prism 4.0 (San Diego, CA):
KNI ¼ exp
mNI½D  DGNI
RT
 
; KIU ¼ exp
mIU½D  DGIU
RT
 
;
YN ¼ aN þ hN D½ ; FrN ¼
1
1 þ KNI þ KNIKIU
;
FrI ¼
KNI
1 þ KNI þ KNIKIU
; FrU ¼
KNIKIU
1 þ KNI þ KNIKIU
;
F ¼ YN þ FrI YI  YNð Þ þ FrU YU  YNð Þ ð5Þ
where YN, YI, and YU are intensities of native, inter-
mediate and unfolded states, respectively. ΔGNI is the
difference in Gibbs' free energy of native and inter-
mediate states, and ΔGIU is the difference between
intermediate and unfolded states. mNI is a constant
related to the average fractional change in degree of
exposure of residues between native and intermediate
states; the same as mIU but between intermediate and
unfolded states. In terms of statistical mechanics, FrN, FrI,
and FrU are fractions of the partition function in a three-
energy-state system, and the plot of fractional popula-
tions of different states versus denaturant concentration
can be generated from these equations. Full datasets
acquired under the same experimental conditions but at
different equilibration times were globally fit to the three-
state model; values for YN, YI, YU and mNI and mIU were
shared between the different equilibration times, whilst
ΔGNI and ΔGNI were allowed to vary. Although there
are a large number of variables in the three-state model
equation, which would make fitting to a single unfolding
curve unreliable, the use of a global analysis allows an
accurate determination of the apparent thermodynamic
parameters.
Solvent-accessible surface area and m-value
calculations
The solvent-accessible surface area (SASA) of native
trGFPuv was calculated from its X-ray crystal
33
Intermediate States in the Unfolding of GFPstructure (PBD code: 1B9C) using the web-based
program GETAREA.39 The SASA of possible intermedi-
ate states were estimated using the following method.First, a partially folded structure of GFP was generated
simply by deletion of different regions from the PDB
file. In generating possible intermediate states we were
guided by the H/D exchange results that showed that
β-strands 7, 8, 9 and 10, have on average higher
exchange rates than the rest of the β-barrel, and may,
therefore, be a cooperatively unfolding unit (see Results
and Discussion). Three possible intermediate states were
considered: I7–10 (β7 to β10 are unfolded), I7–9 (β7 to β9
are unfolded) or I7–8 (β7 to β8 are unfolded). The SASA
of the partially folded structures were then calculated
using GETAREA as described above. Second, the SASA
of the remaining unfolded, unstructured region was
calculated using the Upper Bound Model, which is
based on a hard-sphere simulation.41 This method was
also used for calculating the SASA of the full-length
denatured state assuming that there is no significant
residual structure.
The m-value of a protein is highly correlated to the
change of SASA (ΔSASA) between the different states,42
e.g. native and unfolded state, and the m-value for a
protein without cross-links derived from GdmCl-
induced denaturation curves follows the following
relationship42:
m2value ¼ DSASA  ð0:28F0:03Þ ð6Þ
Theoretical m-values were calculated using this equation
and the ΔSASAs calculated above.
Kinetic measurements
The rate of unfolding of trGFPuv in GdmCl was
measured using stopped-flow apparatus (Applied Photo-
physics). The unfolding buffer and protein solution were
mixed in a 10:1 ratio giving final conditions of 1 μM GFP
in 50 mM Mes (pH 6.0), 0.1 mM TCEP, and GdmCl
concentrations in the range 4.0 M– 7.5 M for the expe-
riments at 25 °C and 1 μM GFP in PBS (pH 7.4), 0.1 mM
TCEP for experiments at 37 °C. Rapidmixingwas achieved
by the simultaneous injection of the solutions through a T-
jet mixing chamber using two Hamilton syringes. The
unfolding of the trGFPuv was monitored by fluorescence
with an excitation wavelength of 395 nm and a 405 nm cut-
off filter. At lower concentrations of GdmCl (2.5M–6.0 M),
manualmixingwas employed and the fluorescence change
monitored using a Varian Cary Eclipse Fluorescence
spectrophotometer. Changes in the green fluorescence
were monitored by excitation at 395 nm and emission at
507 nm. Changes in the tyrosine fluorescence were
monitored by excitation at 275 nm and emission at
303 nm thereby minimising the contribution of the single
tryptophan residue to the spectra. Excitation and emission
band passes were 4 nm. All kinetic traces were averaged
over at least three runs and the data fitted well to an
equation describing a single-exponential process.
Kinetic parameters from equilibrium data
Data were obtained at many different GdmCl
concentrations for the purposes of analyzing the
native-denatured state equilibrium. Fluorescence data
obtained at different equilibration times but at the same
GdmCl concentration were plotted, giving the change
369in fluorescence versus time. These traces were then
fitted to an equation describing a single-exponential
process.

8. Enoki, S., Maki, K., Inobe, T., Takahashi, K., Kamagata,
K., Oroguchi, T. et al. (2006). The equilibriumunfolding
intermediate observed at pH 4 and its relationship
with the kinetic folding intermediates in green
fluorescent protein. J. Mol. Biol. 361, 969–982.
9. Georgescu, J., Rehm, T., Wiehler, J., Steipe, B. & Holak,
T. A. (2003). Backbone H(N), N, C(alpha) and C(beta)
assignment of the GFPuv mutant. J. Biomol. NMR, 25,
161–162.
10. Khan, F., Stott, K. & Jackson, S. (2003). 1H, 15N and 13C
backbone assignment of the green fluorescent protein
(GFP). J. Biomol. NMR, 26, 281–282.
11. Bai, Y., Sosnick, T. R., Mayne, L. & Englander, S. W.
(1995). Protein folding intermediates: native-state
hydrogen exchange. Science, 269, 192–197.
12. Englander, S. W. (2000). Protein folding intermediates
and pathways studied by hydrogen exchange. Annu.
Rev. Biophys. Biomol. Struct. 29, 213–238.
13. Hoang, L., Bedard, S., Krishna, M. M., Lin, Y. &
Englander, S. W. (2002). Cytochrome c folding path-
way: kinetic native-state hydrogen exchange. Proc.
Natl Acad. Sci. USA, 99, 12173–12178.Acknowledgements
The work was funded in part by the Welton
Foundation. J.R.H. acknowledges financial support
from the Cambridge Overseas Trust and J.C. from a
Wellcome Trust International Prize Travelling
Research Fellowship. T.D.C. was funded by the
BBSRC. The authors thank the Biomolecular NMR
Facility (Department of Chemistry, University of
Cambridge), E. Coulstock, and A.L. Mallam for
technical assistance.
Supplementary Data
Supplementary data associated with this article
can be found, in the online version, at doi:10.1016/
j.jmb.2007.04.039
References
1. Tsien, R. Y. (1998). The green fluorescent protein.
Annu. Rev. Biochem. 67, 509–544.
2. Zimmer, M. (2002). Green fluorescent protein (GFP):
applications, structure, and related photophysical
behavior. Chem. Rev. 102, 759–781.
3. Reid, B. G. & Flynn, G. C. (1997). Chromophore
formation in green fluorescent protein. Biochemistry,
36, 6786–6791.
4. Ormo, M., Cubitt, A. B., Kallio, K., Gross, L. A., Tsien,
R. Y. & Remington, S. J. (1996). Crystal structure of the
Aequorea victoria green fluorescent protein. Science,
273, 1392–1395.
5. Yang, F., Moss, L. G. & Phillips, G. N., Jr. (1996). The
molecular structure of green fluorescent protein.
Nature Biotechnol. 14, 1246–1251.
6. Fukuda, H., Arai, M. & Kuwajima, K. (2000). Folding
of green fluorescent protein and the cycle3 mutant.
Biochemistry, 39, 12025–12032.
7. Enoki, S., Saeki, K., Maki, K. & Kuwajima, K. (2004).
Acid denaturation and refolding of green fluorescent
protein. Biochemistry, 43, 14238–14248.
37014. Yan, S., Kennedy, S. D. & Koide, S. (2002). Thermo-
dynamic and kinetic exploration of the energy land-scape of Borrelia burgdorferi OspA by native-state
hydrogen exchange. J. Mol. Biol. 323, 363–375.
15. Dempsey, C. E. (2001). Hydrogen exchange in pep-
tides and proteins using NMR-spectroscopy. Prog.
NMR Spectrosc. 39, 135–170.
16. Krishna, M. M., Hoang, L., Lin, Y. & Englander, S. W.
(2004). Hydrogen exchange methods to study protein
folding. Methods, 34, 51–64.
17. Maity, H., Lim, W. K., Rumbley, J. N. & Englander,
S. W. (2003). Protein hydrogen exchange mechanism:
local fluctuations. Protein Sci. 12, 153–160.
18. Krishna, M. M., Lin, Y., Mayne, L. & Englander, S. W.
(2003). Intimate view of a kinetic protein folding
intermediate: residue-resolved structure, interactions,
stability, folding and unfolding rates, homogeneity.
J. Mol. Biol. 334, 501–513.
19. Jeng, M. F., Englander, S. W., Elove, G. A., Wand, A. J.
& Roder, H. (1990). Structural description of acid-
denatured cytochrome c by hydrogen exchange and
2D NMR. Biochemistry, 29, 10433–10437.
20. Hughson, F. M., Wright, P. E. & Baldwin, R. L. (1990).
Structural characterization of a partly folded apo-
myoglobin intermediate. Science, 249, 1544–1548.
21. Dabora, J. M., Pelton, J. G. & Marqusee, S. (1996).
Structure of the acid state of Escherichia coli ribonu-
clease HI. Biochemistry, 35, 11951–11958.
22. Arai, M. & Kuwajima, K. (1996). Rapid formation of a
molten globule intermediate in refolding of alpha-
lactalbumin. Fold. Des. 1, 275–287.
23. Chamberlain, A. K. & Marqusee, S. (1998). Molten
globule unfolding monitored by hydrogen exchange
in urea. Biochemistry, 37, 1736–1742.
24. Forge, V., Wijesinha, R. T., Balbach, J., Brew, K.,
Robinson, C. V., Redfield, C. & Dobson, C. M. (1999).
Rapid collapse and slow structural reorganisation
during the refolding of bovine alpha-lactalbumin.
J. Mol. Biol. 288, 673–688.
25. Seifert, M. H., Georgescu, J., Ksiazek, D., Smialowski,
P., Rehm, T., Steipe, B. & Holak, T. A. (2003). Backbone
dynamics of green fluorescent protein and the effect of
histidine 148 substitution. Biochemistry, 42, 2500–2512.
26. Sivaraman, T., Arrington, C. B. & Robertson, A. D.
(2001). Kinetics of unfolding and folding from amide
hydrogen exchange in native ubiquitin. Nature Struct.
Biol. 8, 331–333.
27. Benitez-Cardoza, C. G., Stott, K., Hirshberg, M., Went,
H. M., Woolfson, D. N. & Jackson, S. E. (2004).
Exploring sequence/folding space: folding studies on
multiple hydrophobic core mutants of ubiquitin.
Biochemistry, 43, 5195–5203.
28. Arrington, C. B. & Robertson, A. D. (1997). Micro-
second protein folding kinetics from native-state
hydrogen exchange. Biochemistry, 36, 8686–8691.
29. Ferraro, D. M., Lazo, N. D. & Robertson, A. D. (2004).
EX1 hydrogen exchange and protein folding. Biochem-
istry, 43, 587–594.
30. Kiefhaber, T. & Baldwin, R. L. (1995). Kinetics of
hydrogen bond breakage in the process of unfolding
of ribonuclease A measured by pulsed hydrogen
exchange. Proc. Natl Acad. Sci. USA, 92, 2657–2661.
31. Pinheiro, T. J., Cheng, H., Seeholzer, S. H. & Roder, H.
(2000). Direct evidence for the cooperative unfolding
of cytochrome c in lipid membranes from H-(2)H
exchange kinetics. J. Mol. Biol. 303, 617–626.
32. Qu, Y. & Bolen, D. W. (2003). Hydrogen exchange
kinetics of RNase A and the urea:TMAO paradigm.
Biochemistry, 42, 5837–5849.
Intermediate States in the Unfolding of GFP33. Battistutta, R., Negro, A. & Zanotti, G. (2000). Crystal
structure and refolding properties of the mutant

F99S/M153T/V163A of the green fluorescent protein.
Proteins: Struct. funt. Genet. 41, 429–437.
34. Pace, C. N. (1975). The stability of globular proteins.
CRC Crit. Rev. Biochem. 3, 1–43.
35. Stepanenko, O. V., Verkhusha, V. V., Kazakov, V. I.,
Shavlovsky, M. M., Kuznetsova, I. M., Uversky, V. N.
& Turoverov, K. K. (2004). Comparative studies on the
structure and stability of fluorescent proteins EGFP,
zFP506, mRFP1, “dimer2”, and DsRed1. Biochemistry,
43, 14913–14923.
36. Verkhusha, V. V., Kuznetsova, I. M., Stepanenko, O. V.,
Zaraisky, A. G., Shavlovsky, M. M., Turoverov, K. K.
& Uversky, V. N. (2003). High stability of Discosoma
DsRed as compared to Aequorea EGFP. Biochemistry,
42, 7879–7884.
37. Soulages, J. L. (1998). Chemical denaturation: poten-
38. Mayne, L. & Englander, S. W. (2000). Two-state vs.
multistate protein unfolding studied by optical
39. Fraczkiewicz, R. & Braun, W. (1998). Exact and
mechanistic detail in protein engineering analysis.
Curr. Opin. Struct. Biol. 11, 94–100.
45. Sanchez, I. E. & Kiefhaber, T. (2003). Evidence for
sequential barriers and obligatory intermediates in
apparent two-state protein folding. J. Mol. Biol. 325,
367–376.
46. Jackson, S. E. (1998). How do small single-domain
proteins fold? Fold Des. 3, R81–R91.
47. Marley, J., Lu, M. & Bracken, C. (2001). A method
for efficient isotopic labeling of recombinant pro-
teins. J. Biomol. NMR, 20, 71–75.
48. Bodenhausen, G. & Ruben, D. J. (1980). Natural
abundance N-15 NMR by enhanced heteronuclear
spectroscopy. Chem. Phys. Letters, 69, 185–189.
49. Piotto, M., Saudek, V. & Sklenar, V. (1992).
Gradient-tailored excitation for single-quantum
Gradient-tailored water suppression for H-1-N-15
HSQC experiments optimized to retain full sensitivity.
Symposium of Protein Structure. Methuen & Co. Ltd.,
ive
371Intermediate States in the Unfolding of GFPefficient analytical calculation of the accessible
surface areas and their gradients for macromole-
cules. J. Computat. Chem. 19, 319–333.
40. Miller, S., Janin, J., Lesk, A. M. & Chothia, C. (1987).
Interior and surface of monomeric proteins. J .Mol.
Biol. 196, 641–656.
41. Creamer, T. P., Srinivasan, R. & Rose, G. D. (1995).
Modeling unfolded states of peptides and proteins.
Biochemistry, 34, 16245–16250.
42. Myers, J. K., Pace, C. N. & Scholtz, J. M. (1995).
Denaturant m values and heat capacity changes:
relation to changes in accessible surface areas of
protein unfolding. Protein Sci. 4, 2138–2148.
43. Helms, V., Straatsma, T. P. & McCammon, J. A. (1999).
Internal dynamics of green fluorescent protein. J. Phys.
Chem. ser. B, 103, 3263–3269.
44. Oliveberg, M. (2001). Characterisation of the transi-
tion states for protein folding: towards a new level of
(Received 6 November 2006; receLondon.
52. Hvidt, A. & Nielsen, S. O. (1966). Hydrogen exchange
in proteins. Adv. Protein Chem. 21, 287–386.
53. Bai, Y., Milne, J. S., Mayne, L. & Englander, S. W.
(1993). Primary structure effects on peptide group
hydrogen exchange. Proteins: Struct. Funct. Genet. 17,
75–86.
54. Pace, C. N. (1986). Determination and analysis of urea
and guanidine hydrochloride denaturation curves.
Methods Enzymo. 131, 266–280.
55. Santoro, M. M. & Bolen, D. W. (1988). Unfolding free
energy changes determined by the linear extrapola-
tion method. 1. Unfolding of phenylmethanesulfonyl
alpha-chymotrypsin using different denaturants.
Biochemistry, 27, 8063–8068.
56. Barrick, D. & Baldwin, R. L. (1993). Three-state
analysis of sperm whale apomyoglobin folding.
Biochemistry, 32, 3790–3796.
Edited by K. Kuwajima
d in revised form 5 April 2007; accepted 16 April 2007)
Available online 20 April 2007melting and hydrogen exchange. Protein Sci. 9,
1873–1877.
J. Mag. Reson. ser. A, 102, 241–245.
51. Linderstrom-Lang, K. (1958). In (Neuberger, A., ed.),tial impact of undetected intermediates in the free
energy of unfolding and m-values obtained from a
two-state assumption. Biophys. J. 75, 484–492.
NMR-spectroscopy of aqueous-solutions. J. Biomol.
NMR, 2, 661–665.
50. Sklenar, V., Piotto, M., Leppik, R. & Saudek, V. (1993).