RESEARCH ARTICLE | BIOPHYSICS AND COMPUTATIONAL BIOLOGY

CHEMISTRY
OPEN ACCESS

Ligand-­induced protein transition state stabilization switches

the binding pathway from conformational selection to
induced fit
Olof Stenströma,1 , Carl Diehla,1 , Kristofer Modiga , and Mikael Akkea,2

Edited by Lewis Kay, University of Toronto, Toronto, ON, Canada; received October 22, 2023; accepted February 29, 2024

Protein–ligand complex formation is fundamental to biological function. A central

question is whether proteins spontaneously adopt binding-­competent conformations Significance
to which ligands bind conformational selection (CS) or whether ligands induce the
binding-­competent conformation induced fit (IF). Here, we resolve the CS and IF Proteins change their
binding pathways by characterizing protein conformational dynamics over a wide conformations as part of their
range of ligand concentrations using NMR relaxation dispersion. We determined biological functions, for example,
the relative flux through the two pathways using a four-­state binding model that when binding another molecule
includes both CS and IF. Experiments conducted without ligand show that galectin-­3 (ligand). Two pathways are
exchanges between the ground-­state conformation and a high-­energy conformation conceivable: Either the ligand
similar to the ligand-­bound conformation, demonstrating that CS is a plausible path- binds preferentially to a rare
way. Near-­identical crystal structures of the apo and ligand-­bound states suggest that
conformation with higher affinity
the high-­energy conformation in solution corresponds to the apo crystal structure.
(“conformational selection”) or the
Stepwise additions of the ligand lactose induce progressive changes in the relaxation
dispersions that we fit collectively to the four-­state model, yielding all microscopic ligand first binds to the dominant
rate constants and binding affinities. The ligand affinity is higher for the bound-­like conformation and induces a
conformation than for the ground state, as expected for CS. Nonetheless, the IF path- transition to the high-­affinity
way contributes greater than 70% of the total flux even at low ligand concentrations. conformation (“induced fit”). We
The higher flux through the IF pathway is explained by considerably higher rates of determined all rate constants of
exchange between the two protein conformations in the ligand-­associated state. Thus, these two pathways for lactose
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.

the ligand acts to decrease the activation barrier between protein conformations in a binding to the protein galectin-­3.
manner reciprocal to enzymatic transition-­state stabilization of reactions involving Even though galectin-­3 exists in
ligand transformation. a preestablished equilibrium of
protein dynamics | ligand-­binding kinetics | molecular recognition
low-­and high-­affinity
conformations, lactose binds
The mechanism of protein–ligand binding has been central to life science for more than primarily via the induced-­fit
60 y, ever since the seminal papers on enzyme flexibility and activity (1, 2) and allosteric pathway because lactose lowers
regulation (3, 4). At the heart of the matter lies the question whether proteins spontaneously the free-­energy barrier of
adopt conformations complementary to their ligands, to which the ligands subsequently galectin-­3’s conformational
bind, as envisioned by Monod–Wyman–Changeux (4), or whether ligands induce those change. As a result, the reaction
conformations upon binding, as initially proposed by Koshland (1). These mechanisms of flux switches from conformational
molecular recognition are denoted “conformational selection” (CS; sometimes referred to selection to induced fit already at
as “select fit”) and “induced fit” (IF), respectively (Fig. 1A). Koshland-­Nemethy-­Filmer (5) quite low ligand concentrations.
subsequently presented a modified view that allowed for CS as a possible mechanism for
ligand binding. IF has dominated the textbook view of protein–ligand interactions for a
long time. However, research over the past decades increasingly supports CS as an alternative
to the IF paradigm (6, 7). In particular, recent developments in NMR spectroscopy and
computational chemistry have provided evidence of short-­lived, weakly populated con- Author contributions: O.S., C.D., and M.A. designed
formers of the nonbound protein that resemble the ligand-­bound state (8–18). While these research; O.S. and C.D. performed research; K.M. and
M.A. contributed new reagents/analytic tools; O.S., C.D.,
results suggest that ligand binding might proceed via the CS pathway, they do not in K.M., and M.A. analyzed data; and O.S., C.D., K.M., and
themselves provide any direct evidence that this is the case. It is the net reaction flux, rather M.A. wrote the paper.

than the rates and equilibrium populations of individual reaction steps, that determines The authors declare no competing interest.

which pathway is the dominating one (18–22). This article is a PNAS Direct Submission.

NMR relaxation dispersion experiments and line-­shape analysis can probe conforma- Copyright © 2024 the Author(s). Published by PNAS.
This open access article is distributed under Creative
tional transitions that give rise to changes in the chemical shift of individual nuclei, which Commons Attribution License 4.0 (CC BY).
provides a powerful means of monitoring ligand-­binding kinetics under equilibrium 1
Present address: SAR AB, Medicon Village, SE-­223 81
conditions (8, 24–33). The results yield both an atomic-­resolution view of the binding Lund, Sweden.
process, as sensed by the protein, and the microscopic rate constants of the individual 2
To whom correspondence may be addressed. Email:
mikael.akke@bpc.lu.se.
steps of the reaction. Recent work has presented analytical expressions for chemical
This article contains supporting information online at
exchange processes involving an arbitrary number of states (34, 35). Previous studies https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.​
have derived detailed expressions for each of the IF and CS pathways that make it possible 2317747121/-­/DCSupplemental.
to distinguish between these pathways based on the kinetic rate constants (36–42). Published March 25, 2024.

PNAS 2024 Vol. 121 No. 14 e2317747121 https://doi.org/10.1073/pnas.2317747121 1 of 8

 However, it is important to recognize that the two pathways are A
not mutually exclusive but can both contribute simultaneously CS
to the overall reaction flux (41, 43).
Here, we use NMR relaxation dispersion experiments to meas- k12
ure the underlying protein dynamics and determine the relative 1 2
importance the CS and IF pathways for lactose binding to the k21
carbohydrate-­binding domain of human galectin-­3 (galectin-­3C).
Galectin-­3C binds saccharides and synthetic ligands in a shallow, k13 k31 k42 k24
solvent-­exposed groove on one side of its β-­sandwich structure
(23); see Fig. 1B. We performed 15N CPMG relaxation dispersion k34
measurements at variable degrees of ligand saturation and ana- 3 4
lyzed the data using a four-­state model that includes both the CS k43
and IF pathways (Fig. 1A). We found that ligand-­free (apo)
galectin-­3C transiently adopts an “open” conformation similar
to the ligand-­bound state, indicating that the protein is capable IF
of binding ligand via the CS pathway. Indeed, the determined
rate constants reveal that the ligand has a higher affinity for the
open conformation than for the closed one. Yet, the IF pathway
dominates across the entire range of ligand concentrations used B
here, corresponding to saturation levels between 4 and 76%. This
result is explained by significantly higher exchange rates between
states 3 and 4 than between 1 and 2, see Fig. 1A, which implies
that the ligand stabilizes the transition state of the protein con- R144
formational change. H158
W181 N160
Results and Discussion
R162
We investigated the kinetic ligand-­binding mechanisms in galec- N174
tin-­3C by recording 15N CPMG relaxation dispersion experiments
on the apo protein and seven additional samples with increasing E184 E165
Q187
amounts of added ligand; the resulting relaxation rate constants have
been deposited in the Biological Magnetic Resonance Data Bank
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.

(44) under accession numbers 52174 and 52175. By numerically C

fitting the CPMG data globally to a four-­state model, using the
Bloch–McConnell equations (45), we included both the CS and IF
pathways of ligand binding. We analyzed the data in two steps: First,
we fitted the data for the apo state separately using a two-­state model
and then proceeded to fit the concentration dependent data, includ-
ing the results from the apo fit as fixed parameters. Fitting the four-­
state model is tractable in the present case, due in part to the
distinctive variation in dispersion profiles with ligand concentration,
and because one of the four equilibria can be isolated and fully deter-
mined from a two-­state fit of the apo state dynamics, which reduces
the roughness of the target function (46).
Apo Galectin-­3C Transiently Adopts a Conformation Similar
to the Ligand-­Bound Conformation. We first investigated the
conformational dynamics of galectin-­3C in the absence of ligand.
We recorded 15N CPMG relaxation dispersion experiments on apo
galectin-­3C at two static magnetic field strengths, 14.1 T and 18.8
T. Significant exchange contributions to the transverse relaxation
rates were observed for 13 residues in apo galectin-­3C (Fig. 2A and
SI Appendix, Fig. S1A and Table S1; BMRB entry 52174 (47)).
Residue-­specific fits reveal exchange processes on two distinct
timescales. One set of six residues (V189, V213, D215, A216,
L219, and Y221) shows fast exchange dynamics between two
states, kex = (4.8 ± 0.8) × 103 s−1 and pMajor = 0.99 ± 0.01. These
residues are all located in β-­strands 8 to 9 and the connecting
hairpin that form the edge of the β-­sheet on the opposite side of
the protein from the ligand-­binding site (Fig. 2I). Another set
of eight residues, viz. N174, K176, L177, R183, E184, E185,
Q187, and I236 shows slower exchange rates, k12 = 34 ± 4 s−1 and
k21 = 735 ± 80 s−1, and relative populations, p1 = 0.96 ± 0.02 and
p2 = 0.04 ± 0.02. Except for I236, these residues are all located in
Fig. 1.   Ligand binding by galectin-­3C. (A) Schematic outline of the four-­state
model including the conformational-­selection (CS; states 1-­2-­4) and induced-­fit
(IF; states 1-­3-­4) pathways for ligand binding. Free ligand is not shown. State 1
is the binding-­incompetent closed state; state 2 is the binding-­competent open
state; state 3 is the intermediate state in the IF pathway; and state 4 is the
final, highest-­affinity bound state. kon = k13 = k24, koff,3 = k31, and koff,4 = k42. The
macroscopic apo state comprises states 1 and 2, while the macroscopic ligand-­
bound state comprises states 3 and 4. (B) Three-­dimensional structure of lactose-­
bound galectin-­3C. The backbone trace is shown in gray ribbon representation
with ligand-­coordinating side chains highlighted in yellow. The lactose molecule
is shown in stick representation with carbon atoms in green and oxygen atoms in
red. (C) Comparison of the binding site in the apo and lactose-­bound galectin-­3C
crystal structures, highlighting the positions of oxygen atoms in water molecules
(pink spheres) bound to the apo protein and in lactose (red). Water oxygens
primarily match those lactose oxygens that are less exposed to solvent. The
protein backbone and side chains are shown in stick representation: blue, apo;
green, lactose bound. PDB entities: apo, 3zsl; lactose bound, 3zsj (23). Panels (B)
and (C) were prepared using PyMOL (Schrödinger, LLC).
β-­strands 5 to 6 that form part of the binding site; I236 is located
>5 Å from the binding site in β-­strand 11.
The chemical shift difference, Δδ, between exchanging states
can be extracted from the CPMG dispersions (48, 49), see

2 of 8 https://doi.org/10.1073/pnas.2317747121 pnas.org
 A Apo B 4% I
I236

L177
K176

N174
C 8% D 11% R183
E184
E185 Q187
J
E184
L177
R183
Q187 N174

K176
E 21% F 27%
I236
E185

G 30% H 76% V155

L147 R144

K176

N174

R183 E184
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.

Q187

Fig. 2.   Conformational exchange and ligand-­binding dynamics in galectin-­3C. (A–H) Representative 15N CPMG relaxation dispersion data for residue E184.
SI Appendix, Fig. S1 shows the corresponding data for all residues included in the fits. Error bars correspond to ±1 SD. (A) Relaxation dispersions for apo galectin-­3C,
with blue (red) symbols indicating data obtained at a static magnetic field strength of 14.1 T (18.8 T). (B–H) Relaxation dispersions obtained at 14.1 T for galectin-­3C
with varying levels of lactose saturation: (B) 4%, (C) 8 %, (D) 11 %, (E) 21 %, (F) 27 %, (G) 30 %, and (H) 75 %. (I) Residues showing conformational exchange in apo
galectin-­3C mapped onto the structure. Red indicates residues in the binding site showing slower exchange dynamics between states 1 (major population) and
2 (minor population). Blue indicates residues showing fast exchange dynamics unrelated to the exchange between states 1 and 2 (all located on the backside
of the protein, away from the binding site). Side chains involved in ligand coordination are shown in stick representation. (J) Chemical shift differences between
states 1 and 2, ΔδCPMG, in apo galectin-­3C determined from 15N CPMG relaxation dispersion experiments plotted versus chemical shift differences between
the apo and lactose-­bound forms, ΔδHSQC, determined either directly from HSQC spectra of each form (blue symbols; Pearson correlation coefficient: r = 0.96;
slope = 1.82) or from the chemical shifts obtained by fitting the four-­state model (red symbols; r = 0.99; slope = 1.47). A straight line with slope = 1 is drawn to
guide the eye. (K) Residues colored red show conformational exchange in all seven samples with lactose added. Panels (I) and (K) were prepared using PyMOL
(Schrödinger, LLC) and PDB entry 3zsj (23).

SI Appendix, Table S2. Notably, for the eight residues in the bind-
ing site, the chemical shifts of the high-­energy conformation of
apo galectin-­3C are similar to those of lactose-­bound galectin-­3C,
as evidenced by comparing ΔδCPMG obtained from the CPMG
relaxation dispersion data and ΔδHSQC measured from HSQC
spectra of the apo and lactose-­bound states (Fig. 2J); Pearson’s
correlation coefficient r = 0.96. In contrast, none of the exchanging
residues located on the backside of the protein experience any
significant chemical shift perturbation upon lactose binding;
|ΔδHSQC| ≤ 0.025 ppm, further indicating that these residues are
involved in a separate exchange process unrelated to ligand bind-
ing. Note that we do not expect a perfect correlation in Fig. 2J
because minor conformational adjustments upon binding and the
presence of the lactose molecule in the binding site are expected
to give rise to additional chemical shift changes. The good agree-
ment between ΔδCPMG and ΔδHSQC provides strong evidence that
apo galectin-­3C transiently samples conformations resembling
the lactose-­bound state. As shown previously, the crystal structures
of apo and lactose-­bound galectin-­3C are virtually identical (23).
Assuming that lactose-­bound galectin-­3C adopts the same con-
formation in solution as in the crystal, the highly similar chemical
shifts of the apo and lactose-­bound states suggest that the
high-­energy conformation of apo galectin-­3C in solution corre-
sponds to the apo crystal structure. In the apo crystal structure,
several water molecules are present in the ligand binding site with
their oxygen atoms occupying the same positions as oxygens in
the bound lactose. Furthermore, the ligand-­coordinating protein
side chains have the same orientation in the apo and lactose-­bound
crystal structures (Fig. 1C). These observations explain the simi-
larity in chemical shifts (Fig. 2J).
We conducted additional experiments to validate this result and
refute the possibility that there might be residual lactose present
in the apo sample that would result in exchange between a major
population of the apo state and a minor population of the
lactose-­bound state (see SI Appendix for details). Taken together,
our results demonstrate that apo galectin-­3C exchanges between
a “closed” (binding-­incompetent) ground state and an “open”
(binding-­competent) high-­energy state, implying that ligand

PNAS 2024 Vol. 121 No. 14 e2317747121 https://doi.org/10.1073/pnas.2317747121 3 of 8

 binding can occur via conformational selection. Similar results
have been reported for several other proteins (8–11). The exchange
observed for apo galectin-­3C defines the equilibrium between
state 1 (ground state) and state 2 (high-­energy state) of the
four-­state model, K2,1 = k12/k21 = p2/p1. These parameters were
subsequently fixed when fitting the full four-­state model as
described below.
Mapping Binding Pathways by Relaxation Dispersion Measure­
ments at Variable Ligand Concentration. To resolve the CS and IF
binding pathways and determine the flux through each pathway, we
measured 15N CPMG relaxation dispersion on galectin-­3C at seven
different concentrations of lactose, covering saturation levels between
4% and 76%. Five of the eight residues that exhibit exchange in the
absence of lactose also display relaxation dispersions of sufficient
quality at all seven concentrations of lactose: N174, K176, R183,
E184, and Q187. Three additional residues, R144, L147, and
V155, show exchange for all lactose concentrations and were also
included in the analysis (Fig. 2K). The relaxation dispersion data
vary significantly with the concentration of lactose (Fig. 2 B–H and
SI Appendix, Fig. S2; BMRB entry 52175 (50)), as a consequence
of the changing populations and the progressively faster on-­rate.
Fig. 1 defines the four-­state model with the CS and IF pathways
connecting states 1-­2-­4 and 1-­3-­4, respectively. In fitting the
four-­state model, we include three global parameters in addition
to residue-­specific chemical shifts and R2,0 values. The three global
parameters are the on-­rate constant for ligand binding, kon = k13 =
k24, the ratio of the rate constants for ligand release from states 3
and 4, ρoff = koff,3/koff,4, and the ratio of the rate constants for
protein conformational change from open to closed in each path-
way, ρclose = k21/k43. All rate constants (and relative populations)
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.
can be calculated from these parameters and the dissociation con-
stant, Kd, measured by ITC, together with k12 and k21, determined
from CPMG dispersion experiments on the apo state (see Materials
and Methods and SI Appendix for details). The assumption that
the on-­rate constants (k13 and k24) of the two pathways are equal
is expected to hold in the present case for the following reasons.
First, the NMR experiments detect ligand-­induced chemical shift
changes of residues located in the ligand-­binding site. Thus, the
on-­rate constant measured here includes processes such as the
diffusion of the ligand across the protein surface following encoun-
ters where the ligand contacts the protein at locations remote from
the binding site. Second, the open and closed conformations of
galectin-­3C are very similar, as evidenced by the fact that chemical
shift changes between the apo and lactose-­bound states (ΔδHSQC)
are limited and observed only for residues in the binding site. In
the absence of large-­scale conformational changes, it is unlikely
that the on-­rate constant will differ between the two conforma-
tions, especially since lactose binding does not involve interactions
between charged species. The assumption of equal on-­rate con-
stants is not expected to be generally valid for protein–ligand
systems.
To fit the four-­state model to the CPMG dispersion data, we
initially performed grid searches on kon, ρoff, and ρclose, while fitting
the chemical shifts and R20 values, so as to characterize the χ2
space. Projections of the χ2 space onto two dimensions (i.e., pairs
of kon, ρoff, and ρclose; see SI Appendix, Fig. S3) clearly show that
χ2 has a well-­defined minimum for 0.3 < ρclose < 0.4, whereas the
model is less sensitive to ρoff, although there is a preference for
200 < ρoff < 600. Furthermore, we find that low χ2 values are
obtained with 150 × 106 s−1M−1 < kon < 250 × 106 s−1M−1. We
then performed free fits of all parameters, starting from values of
kon, ρoff, and ρclose corresponding to low χ2. The model parameters
were fitted simultaneously to all relaxation data, comprising in
4 of 8 https://doi.org/10.1073/pnas.2317747121
Table 1.   Rate constants describing the four-­state model
Rate constant* Fitted value Estimated error
k12 (s−1) 34 4
−1
k21 (s ) 735 80
kon (106 M−1 s−1) 172 37
koff,3 (105 s−1) 5.5 1.9
koff,4 (103 s−1) 2.1 0.5
3
k34 (10 s ) −1 27 5
3
k43 (10 s ) −1 1.93 0.05
*koff,3, koff,4, k34, and k43 were calculated from the fitted parameters ρoff and ρclose; see Mate-
rials and Methods and SI Appendix for details.
total 8 × 7 dispersion curves, each containing 15 to 24 data points.
The best-­fit results yielded kon = (172 ± 37) × 106 s−1M−1, ρclose=
0.43 ± 0.01, and ρoff = 266 ± 48. From these fitted parameter values,
we then extracted the off-­rate constants, koff,3 = (5.5 ± 1.9) ×
105 s−1 and koff,4 = (2.1 ± 0.5) × 103 s−1, and the rates of confor-
mational exchange involving the macroscopic bound state, k43 =
(1.93 ± 0.05) × 103 s−1 and k34 = (27 ± 5) × 103 s−1. Since k12 and
k21 are known from the experiments conducted on apo galectin-­3C,
described above, the four-­state model is thus completely described
(Table 1).
For comparison with the fitted value of kon, we estimated the
diffusion-­limited on-­rate constant for lactose binding using Eq. 5,
which yielded konD = 10 × 109 s−1M−1. Estimates that take the rel-
ative size of the binding site into account (51, 52) are reduced by
roughly a factor of 10 from this value. Thus, the experimental value
of kon is significantly lower than konD, which is expected because
the former includes not only the diffusion-­limited rate of encounter
but also the diffusive “search” by the ligand across the protein sur-
face from the initial point of contact to the immediate neighbor-
hood of the binding site. This result is also in keeping with our
recent work, showing that several hundred ligand–galectin-­3C
encounters occur for each successful binding event (53).
Ligand Affinity Is Higher for the Open State. The fitted rate
constants enable us to determine the affinity of each protein
conformation (open or closed) for lactose, which is not available
from traditional binding assays, including isothermal titration
calorimetry or fluorescence (54, 55). The rate constants establish
that the lactose affinity is considerably higher for state 2 than for
state 1, Ka,2/Ka,1 = koff,3/koff,4 = 270, in keeping with the notion
that state 2 has an open, binding-­competent conformation,
whereas state 1 is closed and binding-­incompetent, as required
in the CS binding model. These results might give the (incorrect)
impression that the CS pathway dominates—however, at this
level of interpretation, we have not yet considered the critical
determinant, namely the flux through each pathway (19, 20).
The Induced-­Fit Pathway Dominates Despite the Presence of
a Bound-­Like Conformation in the Apo State. Next, we calculated
the fluxes through the CS and IF pathways as a function of lactose
concentration based on the determined rate constants; see Eqs. 6
and 7, which show that the flux through a given pathway is the
inverse of the sum of the “waiting times” at each intermediate
state. The results reveal that the IF pathway dominates under all
conditions used in the present experiments (Fig. 3A). The flux
through the IF pathway varies between 73% of the total flux at
the lowest ligand concentration, [L]tot = 29 µM (4% saturation),
to 99% at the highest ligand concentration, [L]tot = 0.95 mM
(76% saturation). By extrapolation, we find that the binding
pathway switches from CS to IF when the ligand concentration
pnas.org
 A 100
50
0 been reported for the case of coupled folding and binding (41).
B 80
60
40
20
C is conceivable that the relatively weak affinity of lactose leads to the
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.
D samples bound-­like conformations.
Fig. 3.   Fluxes and populations as a function of the degree of saturation
and ligand concentration. (A and B) Fluxes through the CS (blue) and IF (red)
pathways, plotted versus the degree of saturation. (B) Close-­up view of the
flux at low degrees of saturation, showing the cross-­over between the CS and
IF pathways. (C and D) Relative populations of states 1 to 4, plotted versus the
degree of saturation (C), or the free ligand concentration (D). Blue, state 1; red,
state 2; yellow, state 3; and magenta, state 4.
becomes greater than 2 µM, where the relative population of
bound states is only 0.3% (Fig. 3B). This result reinforces the
notion that the existence of exchange between closed (binding-­
incompetent) and open (binding-­competent) conformations
of the apo protein does not in and of itself imply that CS is
the dominant binding pathway (19, 20), even though the open
conformation has a higher affinity for ligand, as shown here.
Rather, the conformational exchange in the free state simply
reflects the fact that the two conformations differ in free energy
by a few kBT and are separated by relatively moderate free energy
barriers that can be readily crossed at ambient temperature, as
expected for a protein that changes its conformation as part of
its biological function.
PNAS 2024 Vol. 121 No. 14 e2317747121
The dominance of the IF binding pathway is explained by a
significantly higher rate of transition from closed to open confor-
mation in the ligand-­bound form, k34 = 27 × 103 s−1, than in the
apo form, k12 = 34 s−1, cf. Fig. 1. Importantly, this result indicates
that the ligand decreases the barrier between the two protein con-
formations (Fig. 4), which provides direct validation of the mech-
anism for ligand-­induced conformational change hypothesized by
Koshland–Nemethy–Filmer (5); similar results have previously
We recognize that this phenomenon is reciprocal to enzymatic
transition-­state stabilization in reactions that change the ligand
conformation or chemical structure. Thus, the effect is fully
expected from a physicochemical perspective, although it is argu-
ably a much less commonly acknowledged concept than enzymatic
transition-­state stabilization.
Fig. 3 C and D show how the relative populations of states 1 to
4 vary with lactose concentration and degree of saturation. State 2
dominates over state 3 in the regime below a protein saturation level
of 42% (Fig. 3C), but again at least 97% of the total flux goes
through the IF pathway under these conditions. The macroscopic
lactose-­bound form is dominated by state 4, but a non-­negligible
fraction of ligand-­bound protein is in state 3, p3:p4 = 0.07:0.93. It
detectable population of closed conformations in the macroscopic
lactose-­bound form, whereas it might be expected that high-­affinity
ligands drive the system further toward state 4 (56). Given that p3:p4
is non-­negligible, we reevaluated the chemical shift comparison in
Fig. 2J by plotting ΔδCPMG (=|δ1 – δ2|) against Δδ14 = |δ1 – δ4|,
instead of ΔδHSQC = |δapo – (p3δ3 + p4δ4)/(p3 + p4)|. The corrected
chemical shift differences are represented by the red symbols in
Fig. 2J, which show improved agreement between the two datasets
(r = 0.99), further bolstering the conclusion that apo galectin-­3C
The Four-­State Model Resolves Conformation-­Specific Chemical
Shifts. The chemical shifts determined for each residue included
in the four-­state model are listed in SI Appendix, Table S3. For
all residues but one, the chemical shifts of states 2 (δ2) and 4 (δ4)
are more similar to one another than they are to either of δ1 or
δ3. These results are perfectly in line with the conformational
changes expected for the CS and IF pathways (cf. Fig. 1) but
are not presumed when constructing the four-­state model. We
note that for all residues but E184, the conformational change
G
state 2
state 1
0
state 3 reaction coordinate
state 4
Fig. 4.   Ligand-­induced transition-­state stabilization of the closed–open
conformational transition. Schematic, qualitative view of the free energy
profile along the reaction coordinate of the closed–open conformational
transitions. The curves illustrate the transition between states 1 and 2 (red,
without ligand) and between states 3 and 4 (blue, with ligand present). The free
energies of the closed states (1 and 3) have been set to the same value (ΔG
= 0) for ease of comparison. The figure is not intended to give a quantitative
representation of the relative free energies of the different states.
https://doi.org/10.1073/pnas.2317747121 5 of 8

CS A dispersion profiles maintain a noticeable upward bend to higher
R2 (1/s)
IF B lactose binding to galectin-­3C indicate that exchange between
R2 (1/s)
4-state C appreciably to the overall flux (cf. Fig. 1A).
R2 (1/s)
cp (Hz)
Fig. 5.   Relaxation dispersion profiles as a function of ligand concentration for
different binding mechanisms. (A) CS binding pathway. (B) IF binding pathway.
(C) The four-­state binding model comprising both CS and IF. The individual
curves show relaxation dispersions for protein samples with different extents
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.
of ligand saturation, indicated by the color scheme going from light blue for
0% saturation to magenta for 90% saturation, in steps of 10%; in total 10
curves. In panel (A), with increasing ligand saturation, the dispersion curves
reach higher R2 values at νcp = 0 and show a sharper decline at higher νcp. In
panel (B), the dispersion profile is flat at 0% saturation, then first increases
with increasing saturation, followed by a decrease to again reach a flat plateau.
In panel (C), the dispersion profiles resemble those in panel (B) but start out
with a significant dispersion at 0% saturation [identical to the case in panel
(A)]. The figure was generated by simulating relaxation dispersions for the
different binding mechanisms using the rate constants determined herein
for lactose binding to galectin-­3C.
from state 3 to 4 results in a greater change in chemical shift
than does binding of lactose to the open conformation (i.e., the
change from state 2 to 4). This result might also be expected since
the chemical shift is arguably more sensitive to local and distinct
conformational changes than to low-­affinity binding of lactose
onto the closed conformation. Thus, the overall picture emerging
from the chemical shifts extracted by fitting the four-­state model
makes intuitive sense.
Relaxation Dispersion Profiles Are Diagnostic of the Binding
Mechanism. Fig. 5 compares relaxation dispersion curves for
the full four-­state model with those expected for purely CS or
IF binding mechanisms, using for each pathway the relevant
exchange parameters resulting from the fit of the four-­state
model to our experimental data. Clearly, the dispersion curves
describing the four-­state model, which is a representation of the
underlying experimental data, are very different from the pure
CS or IF scenarios. The simulated CPMG relaxation dispersion
profiles reveal diagnostic signatures for distinguishing between the
different reaction mechanisms, as has recently been demonstrated
for the IF and CS pathways by the Weikl and Griesinger groups
(42). For example, when the CS pathway dominates, R2 decreases
6 of 8 https://doi.org/10.1073/pnas.2317747121
progressively with increasing lactose concentration, but the
R2 values at low νcp, even for the highest [L] (Fig. 5A). By contrast,
when the IF pathway dominates, R2 starts out at low values and
first increases progressively with increasing [L], while flattening
out at high [L] (Fig. 5B). The experimental data are effectively a
mixture of the dispersion profiles expected for pure IF and CS
binding pathways, in that the profiles progressively decrease in R2
and flatten out with increasing [L], without showing any upward
bend at low νcp (Fig. 5C). Thus, the dispersion profiles describing
all 4 states must be taken into account, even though the flux
through the CS pathway is negligible at higher [L]. This result
is explained by the fact that the exchange between states 1 and 2
(a key feature of the CS model) exist in conjunction with the IF
pathway. That is, in the case of lactose binding to galectin-­3C,
the four-­state exchange predominantly follows a linear exchange
mechanism: 2 ⇌ 1 ⇌ 3 ⇌ 4, where state 1 is dominant at low
ligand concentrations, whereas the step 2 ⇌ 4 does not contribute
Concluding Remarks. We have shown that protein NMR relaxation
dispersion experiments can resolve the CS and IF ligand-­binding
pathways, even in those cases where both contribute to binding. By
fitting a four-­state binding model comprising both the CS and IF
pathways, we determined the individual rate constants that provide
information on conformation-­specific ligand affinities, as well as
the relative flux through the two pathways. Our results reveal that
galectin-­3C binds lactose almost exclusively via the IF pathway,
even though the apo protein populates bound-­like conformations
with higher ligand affinity. The dominance of the IF pathway arises
primarily from a faster rate of protein conformational change in
the lactose-­associated form than in the apo form. Thus, the ligand
acts to lower the transition barrier for protein conformational
dynamics, a phenomenon reciprocal to enzymatic transition-­state
stabilization. Our results provide an enriched view of induced fit
and conformational selection that extends current understanding of
protein–ligand binding processes.
Materials and Methods
15
Protein Expression and Sample Preparation. N-­labeled galectin-­3C sam-
ples were expressed, purified, and prepared as described (54, 57). Following
an initial set of CPMG experiments on the original apo galectin-­3C sample, the
sample was filtered using a Vivaspin 2 MWCO 5,000 centrifugal concentrator to
dilute by a factor of 104 any lactose possibly remaining in the original sample.
Lactose titrations were carried out by titrating small aliquots of lactose (8.8 mM or
35 mM lactose dissolved in NMR buffer) into three different galectin-­3C samples,
resulting in seven samples covering 7 to 75% saturation of the protein with ligand
(see SI Appendix, Table S4 for detailed information on sample concentrations).
15
N CPMG relaxation dispersion datasets were acquired at each titration point.
Relaxation Dispersion Experiments. Relaxation-­compensated CPMG 15N
relaxation dispersion experiments (58) were performed at static magnetic field
strengths of 14.1 T (apo and lactose-­containing samples) and 18.8 T (apo sample
only) and a temperature of 301 ± 0.1 K, which was calibrated using a methanol
sample before each series of experiments. An equilibration period of 5 ms was
added prior to the CPMG refocusing pulse train to ensure that the initial magnet-
izations reflect the relative populations of the exchanging states (59). The experi-
ments performed at 18.8 T utilized the phase cycle proposed by Yip and Zuiderweg
to suppress artifacts due to off-­resonance effects (60). The effective relaxation rate,
R2, at a given CPMG refocusing frequency was determined from 2 data points (61).
Relaxation dispersions were sampled using 18 to 24 refocusing frequencies, νcp.
The sign of the chemical shift difference between the exchanging conformations in
the apo state was determined using the HSQC/HMQC approach (62). The relaxation
pnas.org
 data have been deposited in the Biological Magnetic Resonance Data Bank (BMRB)
as entries 52174 (apo) (47) and 52175 (lactose, all concentrations) (50): DOI:
10.13018/BMR52174; DOI: 10.13018/BMR52175.
Data Analysis. NMR data were processed using NMRPipe (63). The processing
protocol involved a solvent filter, cosine-­squared apodization function, zero filling
to twice the number of points in both dimensions, and a polynomial baseline
correction in the 1H dimension. Peak integration and error analysis was performed
using PINT (64, 65). Errors in the extracted R2 values were estimated from the
S/N using error propagation.
The relaxation dispersion data of apo galectin-­3C were analyzed using CPMGfit
v2.42, an in-­house Matlab program. The exchange parameters of the Carver-­Richards
two-­state exchange model were fitted to the relaxation dispersion data (24, 66):
R2 + R20 + Rex (1 ∕ 𝜏),
in which
{ }
1 1 −1
( ( ) ( ))
Rex (1 ∕ 𝜏) = kex − cosh D + cosh 𝜂 + − D _ cosh 𝜂 − ,
2 𝜏
1
{ )1∕2 }
𝜓 + 2Δ𝜔2 + 𝜓 2 + 𝜁 2
( ) (
D± = ±1+ ,
2
𝜏
�� �1∕2 ��1∕2
±𝜓 + 𝜓2 +𝜁2
�
𝜂± = √ , [4]
2 bound complex, NA is Avogadro’s number, and D is the sum of the diffusion
and ψ = kex2 – Δω2, ζ = –2Δωkex(1–2pminor), kex = k1 + k−1 is the sum of the
forward and reverse rate constants, Δω = Δδ γ B0 is the chemical shift difference
between the exchanging conformations, where γ is the gyromagnetic ratio of the
nuclear spin and B0 is the static magnetic field strength, R20 is the average limiting
value of the relaxation rate constant for processes other than chemical exchange,
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.
pminor is the population of the minor (less populated) conformational state, which
is related to the major conformational state by pmajor = 1−pminor, and τ = 1/(2νcp)
is the spacing between refocusing pulses in the CPMG train.
Relaxation dispersion data for apo galectin-­3C were initially modeled by
residue-­specific fits of the Carver–Richards equations. The statistical significance
of each fit was assessed by also fitting the data to a constant R20 value (i.e., mod-
eling a flat relaxation profile, indicating the absence of exchange), and the F-­test
was used to discriminate between models by rejecting the simpler model at the
level P < 0.01 (67). In a subsequent step, residues with similar residue-­specific
exchange parameters were fitted simultaneously to yield a single, global set of
parameters kex and pminor. Errors in the fitted parameters were estimated from
1,000 synthetic datasets created using Monte–Carlo simulations (68).
To fit the 15N CPMG relaxation dispersions of galectin-­3C with variable
amounts of lactose present, we constructed a four-­state model based on the
Bloch–McConnell equation (46), in which state 1 is the ligand-­free, closed ground
state; state 2 is the ligand-­free, open high-­energy state resembling the ligand-­
bound state; state 3 is the ligand-­associated, closed state; and state 4 is the
ligand-­bound, open state (Fig. 1). This model includes both the CS (states 1-­2-­4)
and IF (states 1-­3-­4) pathways. The exchange parameters determined from the
experiments on apo galectin-­3C, i.e., k12 and k21, were fixed when fitting the four-­
state model. δ1 was fixed to the value measured in the 15N HSQC spectrum of apo
galectin-­3C (given that p2 ≪ p1, the contribution from δ2 on the apo chemical
shift is insignificant). δ2 and δ3 were fitted while δ4 was calculated from p3, p4,
and the chemical shift measured in the 15N HSQC spectrum of lactose-­saturated
galectin-­3C: δlac = p3δ3 + p4δ4. R20 was fitted individually for each residue, but
globally over all lactose concentrations.
We introduced two parameters ρoff and ρclose describing the ratios of pairs
of corresponding rates along the two pathways. ρoff describes the ratio of the
1. D. E. Koshland, Application of the theory of enzyme specificity to protein synthesis. Proc. Natl. Acad.
Sci. U.S.A. 44, 98–104 (1958).
2. D. E. Koshland, Enzyme flexibility and enzyme action. J. Cell Comp. Physiol. 54, 245–258
(1959).
PNAS 2024 Vol. 121 No. 14 e2317747121
ligand off-­rate constants from states 3 and 4, i.e., ρoff = koff,3/koff,4, where koff,3 and
koff,4 relate to the IF and CS pathway, respectively. ρclose describes the ratio of the
rates going from the open conformation to the closed conformation along the
CS and IF pathways, i.e., ρclose = k21/k34 (Fig. 1). Furthermore, we assumed that
the on-­rate constant of lactose is the same for the CS and IF pathways, which is
reasonable given that the NMR relaxation experiment reports on chemical shift
changes between states 1 and 3, or 2 and 4, induced by ligand binding to the
binding site or its proximity.
Prior to minimization of the target function, we performed extensive grid
searches on the parameters ρoff, ranging from 30 to 1,000; ρclose, 0.001 to
1,000; and kon, (8 to 300) × 106 M−1 s−1, while the chemical shift differences
and R20 were optimized. The combination of parameters yielding the lowest
χ2 was taken as input for the minimization algorithm. At each iteration, the full
relaxation dispersion curve was simulated for all residues, using the current
[1] values of the fitted model parameters in the Bloch–McConnell equations, and
χ2 was evaluated. Errors in the fitted parameters were estimated from Monte–
Carlo simulations using 500 synthetic datasets normally distributed by the
experimental SDs (68). The Matlab code for fitting the relaxation dispersion data
[2]
is available on GitHub (https://github.com/Akke-­group/4 _ state_fits.git) (69).
We estimated the diffusion-­limited on-­rate constant for lactose binding based
on the relationship (21):
[3]
kon,D = 4𝜋 r DNA × 103 , [5]
where r ≈ 20 Å is the distance between the centers of ligand and protein in the
constants for lactose, Dlac = 0.57 10−9 m2s−1 (70), and galectin-­3C, DP = 0.126
10−9 m2s−1 as calculated using HydroNMR (71).
The fluxes through the CS pathway (FCS) and the IF pathway (FIF) were calcu-
lated as (20)
)−1
(
1 1
FCS = [ ] + [ ][ ] , [6]
k12 P1 k24 P2 L
)−1
(
1 1
FIF = [ ][ ] + [ ] , [7]
k13 P1 L k34 P3 L
where [P1] and [P2] are the protein concentrations of states 1 and 2, respectively, [L] is
the free ligand concentration, [P3L] is the concentration of the protein–ligand complex
in state 3, and kAB is the rate constant going from state A to B. Thus, the flux through a
given pathway is the inverse of the sum of the waiting times at each intermediate state.
Data, Materials, and Software Availability. NMR relaxation rate constants
data have been deposited in BMRB (52174 (47) and 52175 (50)). The Matlab
code for fitting the relaxation dispersion data is available on GitHub (https://
github.com/Akke-­group/4_state_fits.git) (69).
ACKNOWLEDGMENTS. We thank Kristine Steen Jensen for helpful discussions
and an anonymous reviewer for valuable comments. High-­field (18.8 T) NMR
data were acquired at the Swedish NMR Center at the University of Gothenburg.
This work was funded by the Swedish Research Council (2018-­4995), the FLÄK
Research School for Pharmaceutical Sciences at Lund University, the Knut and
Alice Wallenberg Foundation (2013.0022), and the European Union (ERC,
DynaPLIX, SyG-­2022 101071843). Views and opinions expressed are however
those of the authors only and do not necessarily reflect those of the European
Union or the European Research Council. Neither the European Union nor the
granting authority can be held responsible for them.
Author affiliations: aDivision of Biophysical Chemistry, Center for Molecular Protein
Science, Department of Chemistry, Lund University, SE-­221 00 Lund, Sweden
3. J. Monod, J. P. Changeux, F. Jacob, Allosteric proteins and cellular control systems. J. Mol. Biol. 6,
306–329 (1963).
4. J. Monod, J. Wyman, J.-­P. Changeux, On the nature of allosteric transitions: A plausible model. J.
Mol. Biol. 12, 88–118 (1965).
https://doi.org/10.1073/pnas.2317747121 7 of 8
 5. D. E. Koshland, G. Nemethy, D. Filmer, Comparison of experimental binding data and theoretical
models in proteins containing subunits. Biochemistry 5, 365–385 (1966).
6. J. P. Changeux, S. Edelstein, Conformational selection or induced fit? 50 years of debate resolved.
F1000 Biol. Rep. 3, 19 (2011).
7. D. D. Boehr, R. Nussinov, P. E. Wright, The role of dynamic conformational ensembles in
biomolecular recognition. Nat. Chem. Biol. 5, 789–796 (2009).
8. A. Malmendal, J. Evenäs, S. Forsén, M. Akke, Structural dynamics in the C-­terminal domain of
calmodulin at low calcium levels. J. Mol. Biol. 293, 883–899 (1999).
9. E. Z. Eisenmesser et al., Intrinsic dynamics of an enzyme underlies catalysis. Nature 438, 117–121
(2005).
10. D. D. Boehr, D. McElheny, H. J. Dyson, P. E. Wright, The dynamic energy landscape of dihydrofolate
reductase catalysis. Science 313, 1638–1642 (2006).
11. H. Beach, R. Cole, M. L. Gill, J. P. Loria, Conservation of us-­ms enzyme motions in the apo-­and
substrate-­mimicked state. J. Am. Chem. Soc. 127, 9167–9176 (2005).
12. K. Okazaki, S. Takada, Dynamic energy landscape view of coupled binding and protein
conformational change: Induced-­fit versus population-­shift mechanisms. Proc. Natl. Acad. Sci. U.S.A.
105, 11182–11187 (2008).
13. O. F. Lange et al., Recognition dynamics up to microseconds revealed from an RDC-­derived
ubiquitin ensemble in solution. Science 320, 1471–1475 (2008).
14. S.-­R. Tzeng, C. G. Kalodimos, Dynamic activation of an allosteric regulatory protein. Nature 462,
368–372 (2009).
15. N. J. Anthis, M. Doucleff, G. M. Clore, Transient, sparsely populated compact states of apo and
calcium-­loaded calmodulin probed by paramagnetic relaxation enhancement: Interplay of
conformational selection and induced fit. J. Am. Chem. Soc. 133, 18966–18974 (2011).
16. L. Ye, N. Van Eps, M. Zimmer, O. P. Ernst, R. Scott Prosser, Activation of the A2A adenosine G-­protein-­
coupled receptor by conformational selection. Nature 533, 265–268 (2016).
17. V. Venditti, G. M. Clore, Conformational selection and substrate binding regulate the monomer/
dimer equilibrium of the C-­terminal domain of Escherichia coli enzyme I. J. Biol. Chem. 287,
26989–26998 (2012).
18. A. Sekhar et al., Conserved conformational selection mechanism of Hsp70 chaperone-­substrate
interactions. eLife 7, e32764 (2018).
19. V. Z. Miloushev et al., Dynamic properties of a type II cadherin adhesive domain: Implications for the
mechanism of strand-­swapping of classical cadherins. Structure 16, 1195–1205 (2008).
20. G. G. Hammes, Y. C. Chang, T. G. Oas, Conformational selection or induced fit: A flux description of
reaction mechanism. Proc. Natl. Acad. Sci. U.S.A. 106, 13737–13741 (2009).
21. K. S. Chakrabarti et al., Conformational selection in a protein-­protein interaction revealed by
dynamic pathway analysis. Cell Rep. 14, 32–42 (2016).
22. T. R. Weikl, F. Paul, Conformational selection in protein binding and function. Protein Sci. 23,
1508–1518 (2014).
23. K. Saraboji et al., The carbohydrate-­binding site in galectin-­3 is preorganized to recognize a sugar-­
like framework of oxygens: Ultra-­high resolution structures and water dynamics. Biochemistry 51,
296–306 (2012).
24. D. G. Davis, M. E. Perlman, R. E. London, Direct measurements of the dissociation-­rate constant
Downloaded from https://www.pnas.org by 102.188.205.138 on April 3, 2024 from IP address 102.188.205.138.
for inhibitor-­enzyme complexes via the T1r and T2 (CPMG) methods. J. Magn. Reson. Ser. B 104,
266–275 (1994).
25. R. Schneider et al., Visualizing the molecular recognition trajectory of an intrinsically disordered
protein using multinuclear relaxation dispersion NMR. J. Am. Chem. Soc. 137, 1220–1229 (2015).
26. O. Stenström, C. Diehl, K. Modig, U. J. Nilsson, M. Akke, Mapping the energy landscape of
protein–ligand binding via linear free energy relationships determined by protein NMR relaxation
dispersion. RSC Chem. Biol. 2, 259–265 (2021).
27. V. A. Feher, E. P. Baldwin, F. W. Dahlquist, Access of ligands to cavities within the core of a protein is
rapid. Nat. Struct. Biol. 3, 516–521 (1996).
28. D. Zeng, V. S. Bhatt, Q. Shen, J. H. Cho, Kinetic insights into the binding between the nSH3 domain
of CrkII and proline-­rich motifs in cAbl. Biophys. J.111, 1843–1853 (2016).
29. E. Z. Eisenmesser, D. A. Bosco, M. Akke, D. Kern, Enzyme dynamics during catalysis. Science 295,
1520–1523 (2002).
30. K. Sugase, J. C. Lansing, H. J. Dyson, P. E. Wright, Tailoring relaxation dispersion experiments for
fast-­associating protein complexes. J. Am. Chem. Soc. 129, 13406–13407 (2007).
31. D. Tolkatchev, P. Xu, F. Ni, Probing the kinetic landscape of transient peptide-­protein interactions by
use of peptide 15N NMR relaxation dispersion spectroscopy: Binding of an antithrombin peptide to
human prothrombin. J. Am. Chem. Soc. 125, 12432–12442 (2003).
32. S. De et al., Complete thermodynamic and kinetic characterization of the isomer-­specific interaction
between Pin1-­WW domain and the amyloid precursor protein cytoplasmic tail phosphorylated at
Thr668. Biochemistry 51, 8583–8596 (2012).
33. A. Furukawa, T. Konuma, S. Yanaka, K. Sugase, Quantitative analysis of protein–ligand interactions by
NMR. Prog. Nucl. Magn. Reson. Spectrosc. 96, 47–57 (2016).
34. H. Koss, M. Rance, A. G. Palmer, General expressions for Carr-­Purcell-­Meiboom-­Gill relaxation
dispersion for N-­site chemical exchange. Biochemistry 57, 4753–4763 (2018).
35. H. Koss, M. Rance, A. G. Palmer, Algebraic expressions for Carr-­Purcell-­Meiboom-­Gill relaxation
dispersion for N-­site chemical exchange. J. Magn. Reson. 321, 106846 (2020).
36. A. D. Vogt, E. Di Cera, Conformational selection or induced fit? A critical appraisal of the kinetic
mechanism. Biochemistry 51, 5894–5902 (2012).
37. A. D. Vogt, E. Di Cera, Conformational selection is a dominant mechanism of ligand binding.
Biochemistry 52, 5723–5729 (2013).
38. F. Paul, T. R. Weikl, How to distinguish conformational selection and induced fit based on chemical
relaxation rates. PLoS Comput. Biol. 12, e1005067 (2016).
8 of 8 https://doi.org/10.1073/pnas.2317747121
39. S. Gianni, J. Dogan, P. Jemth, Distinguishing induced fit from conformational selection. Biophys.
Chem. 189, 33–39 (2014).
40. T. R. Weikl, D. D. Boehr, Conformational selection and induced changes along the catalytic cycle of
Escherichia coli dihydrofolate reductase. Proteins 80, 2369–2383 (2012).
41. K. G. Daniels, Y. Suo, T. G. Oas, Conformational kinetics reveals affinities of protein conformational
states. Proc. Natl. Acad. Sci. U.S.A. 112, 9352–9357 (2015).
42. K. S. Chakrabarti et al., A litmus test for classifying recognition mechanisms of transiently binding
proteins. Nat. Commun. 13, 3792 (2022).
43. A. G. Palmer, Chemical exchange in biomacromolecules: Past, present, and future. J. Magn. Reson.
241, 3–17 (2014).
44. J. C. Hoch et al., Biological magnetic resonance data bank. Nucleic Acids Res. 51, D368–D376 (2023).
45. H. M. McConnell, Reaction rates by nuclear magnetic resonance. J. Chem. Phys. 28, 430–431
(1958).
46. P. Li, I. R. S. Martins, M. K. Rosen, The feasibility of parameterizing four-­state equilibria using
relaxation dispersion measurements. J. Biomol. NMR 51, 57–70 (2011).
47. O. Stenström, C. Diehl, K. Modig, M. Akke, Ligand-­induced transition state stabilization of protein
conformational change switches the binding pathway from conformational selection to induced
fit -­APO. Biological Magnetic Resonance Data Bank. https://doi.org/doi:10.13018/BMR52174.
Deposited 15 October 2023.
48. E. L. Kovrigin, J. G. Kempf, M. J. Grey, J. P. Loria, Faithful estimation of dynamics parameters from
CPMG relaxation dispersion measurements. J. Magn. Reson. 179, 178–189 (2006).
49. A. G. Palmer, C. D. Kroenke, J. P. Loria, Nuclear magnetic resonance methods for quantifying
microsecond-­to-­millisecond motions in biological macromolecules. Methods Enzymol. 339,
204–238 (2001).
50. O. Stenström, C. Diehl, K. Modig, M. Akke, Ligand-­induced transition state stabilization of protein
conformational change switches the binding pathway from conformational selection to induced fit
-­LACTOSE. Biological Magnetic Resonance Data Bank. https://doi.org/doi:10.13018/BMR52175.
Deposited 15 October 2023.
51. G. Schreiber, G. Haran, H. X. Zhou, Fundamental aspects of protein -­Protein association kinetics.
Chem. Rev. 109, 839–860 (2009).
52. O. G. Berg, Orientation constraints in diffusion-­limited macromolecular association. The role of
surface diffusion as a rate-­enhancing mechanism. Biophys. J. 47, 1–14 (1985).
53. S. Wernersson, S. Birgersson, M. Akke, Cosolvent dimethyl sulfoxide influences protein-­ligand
binding kinetics via solvent viscosity effects: Revealing the success rate of complex formation
following diffusive protein-­ligand encounter. Biochemistry 62, 44–52 (2023).
54. C. Diehl et al., Protein flexibility and conformational entropy in ligand design targeting the
carbohydrate recognition domain of galectin-­3. J. Am. Chem. Soc. 132, 14577–14589 (2010).
55. M. L. Verteramo et al., Interplay between conformational entropy and solvation entropy in protein-­
ligand binding. J. Am. Chem. Soc. 141, 2012–2026 (2019).
56. M. Ubbink, The courtship of proteins: Understanding the encounter complex. FEBS Lett. 583,
1060–1066 (2009).
57. C. Diehl, S. Genheden, K. Modig, U. Ryde, M. Akke, Conformational entropy changes upon lactose
binding to the carbohydrate recognition domain of galectin-­3. J. Biomol. NMR 45, 157–169 (2009).
58. J. P. Loria, M. Rance, A. G. Palmer, A relaxation-­compensated Carr-­Purcell-­Meiboom-­Gill sequence for
characterizing chemical exchange by NMR spectroscopy. J. Am. Chem. Soc. 121, 2331–2332 (1999).
59. D. F. Hansen, P. Vallurupalli, L. E. Kay, An improved N-­15 relaxation dispersion experiment for the
measurement of millisecond time-­scale dynamics in proteins. J. Phys. Chem. B 112, 5898–5904
(2008).
60. G. N. B. Yip, E. R. P. Zuiderweg, A phase cycle scheme that significantly suppresses offset-­dependent
artifacts in the R-­2-­CPMG N-­15 relaxation experiment. J. Magn. Reson. 171, 25–36 (2004).
61. F. A. A. Mulder, N. R. Skrynnikov, B. Hon, F. W. Dahlquist, L. E. Kay, Measurement of slow
(microseconds-­milliseconds) time scale dynamics in protein side chains by 15N relaxation
dispersion NMR spectroscopy: Application to Asn and Gln residues in a cavity mutant of T4
lysozyme. J. Am. Chem. Soc. 123, 967–975 (2001).
62. N. R. Skrynnikov, F. W. Dahlquist, L. E. Kay, Reconstructing NMR spectra of “invisible” excited states
using HSQC and HMQC experiments. J. Am. Chem. Soc. 124, 12352–12360 (2002).
63. F. Delaglio et al., NMRPipe: A multidimensional spectral processing system based on UNIX pipes.
J. Biomol. NMR 6, 277–293 (1995).
64. A. Ahlner, M. Carlsson, B. H. Jonsson, P. Lundström, PINT: A software for integration of peak volumes
and extraction of relaxation rates. J. Biomol. NMR 56, 191–202 (2013).
65. M. Niklasson et al., Comprehensive analysis of NMR data using advanced line shape fitting.
J. Biomol. NMR 69, 93–99 (2017).
66. J. P. Carver, R. E. Richards, A general two-­site solution for the chemical exchange produced
dependence of T2 upon the Carr-­Purcell pulse separation. J. Magn. Reson. 6, 89–105 (1972).
67. J. L. Devore, Probability and Statistics for Engineering and the Sciences (Brooks/Cole Publishing
Company, ed. 5, 1999).
68. W. H. Press, B. P. Flannery, S. A. Teukolsky, W. T. Vetterling, Numerical Recipes. The Art of Scientific
Computing (Cambridge University Press, 1986).
69. O. Stenström, K. Modig, M. Akke, Matlab scripts to fit ligand-­concentration dependent CPMG
relaxation dispersion data to a 4-­state exchange binding model. GitHub. https://github.com/Akke-­
group/4_state_fits. Accessed 20 February 2024.
70. A. C. F. Ribeiro et al., Binary mutual diffusion coefficients of aqueous solutions of sucrose, lactose,
glucose, and fructose in the temperature range from (298.15 to 328.15) K. J. Chem. Eng. Data 51,
1836–1840 (2006).
71. P. Bernado, J. Garcia de la Torre, M. Pons, Interpretation of 15N NMR relaxation data of globular
proteins using hydrodynamic calculations with HYDRONMR. J. Biomol. NMR 23, 139–150 (2002).
pnas.org