Dissecting the hydrophobic effect on the molecular level: The role of water, enthalpy, and entropy in ligand binding to thermolysin

Article abstract of DOI:10.1002/anie.201208561

The hydrophobic effect is associated with the successive replacement of water molecules in the binding site of a protein by hydrophobic groups of the ligand. Although the hydrophobic effect is assumed to be entropy-driven, large changes in enthalpy and entropy are observed with the model system thermolysin. Structural changes in the binding features of the water molecules ultimately determine the thermodynamics of the hydrophobic effect. Copyright

Full text of DOI:10.1002/anie.201208561

.
Angewandte
Communications
DOI: 10.1002/anie.201208561
Protein–Ligand Complexes
Dissecting the Hydrophobic Effect on the Molecular Level: The Role
of Water, Enthalpy, and Entropy in Ligand Binding to Thermolysin**
Adam Biela, Nader N. Nasief, Michael Betz, Andreas Heine, David Hangauer,* and
Gerhard Klebe*
Dedicated to Professor Jack D. Dunitz on the occasion of his 90th birthday
The hydrophobic effect is viewed as the driving force for the
aggregation of nonpolar substances with extended lipophilic
molecular surfaces in aqueous solution through the exclusion
of water molecules from the formed interfaces.[1,2] It is usually
quoted to explain why an oil/water mixture spontaneously
separates, why soluble proteins fold with a hydrophobic core
and a hydrophilic outer surface,[3,4] why membrane compo-
nents assemble as lipid bilayers and micelles, why membrane
proteins are accommodated in membrane segments, and why
small molecules associate in protein binding pockets with
mutual burial of hydrophobic surfaces.[5] In the latter instance,
it is a general strategy in medicinal chemistry to improve
protein–ligand binding by increasing the ligandꢀs hydrophobic
surface which becomes buried in hydrophobic pockets of the
target protein. In all cases, the hydrophobic effect is
considered to be the major force of association. On the
molecular level, this phenomenon is commonly attributed to
the displacement of water molecules arranged around the
hydrophobic surfaces, and entropic effects are made respon-
sible to drive this association. The entropic profile is related to
changes in the degree of ordering and the dynamic properties
of the water molecules, which are assumed to be more
disordered in the bulk water phase relative to where they
were located prior to being displaced upon hydrophobic
association. Recent studies have demonstrated, however, that
hydrophobic interactions can originate either from enthalpy-
or entropy-driven binding, making simple explanations often
presented for the hydrophobic effect insufficient.[6–12] Also in
computational design tools the handling of explicit water
molecules has received increasing recognition. Tools such as
WaterMap and Szmap[13,14] try to take into account water
structures in drug design and the properties of individual
water molecules are discussed in terms of enthalpy and
entropy.
In order to obtain a better understanding of the hydro-
phobic effect on the molecular level and its role in protein–
ligand binding, we embarked on a systematic study using
thermolysin (TLN) as a model system.[6] This thermostable
bacterial zinc metalloprotease from Bacillus thermoproteoly-
ticus exhibits three specificity pockets of predominantly
hydrophobic nature (Scheme 1). It has been considered
[*] Dr. A. Biela, M. Betz, Dr. A. Heine, Prof. Dr. G. Klebe
Department of Pharmaceutical Chemistry
Philipps-University Marburg
Scheme 1. A schematic view of the binding pocket of thermolysin with
the bound ligand and the substitution pattern of the studied ligands
1–8.
Marbacher Weg 6, 35032 Marburg (Germany)
E-mail: klebe@mailer.uni-marburg.de
N. N. Nasief, Prof. Dr. D. Hangauer
Department of Chemistry, University at Buffalo
The State University of New York
Buffalo, NY 14260 (USA)
a prototype for the entire class of enzymes[15] owing to its
highly conserved active-site architecture, despite remarkable
sequence differences to other zinc proteases. Potent TLN
inhibitors are often designed as transition-state ana-
logues.[16–18] The enzyme has been frequently used as a surro-
gate[19–23] for other metalloenzymes against which new drugs
are developed, and served as a model system to test ideas[24,25]
and new methodological concepts.[26] TLN was one of the first
crystallographically investigated metalloproteases[27,28] and its
catalytic zinc ion is coordinated by His142, His146, and
Glu166. The adjacent S1 subsite is rather nonspecific and
accommodates hydrophobic ligand portions.[16] In contrast,
the S1’ pocket is a deep and well-defined cavity which hosts
E-mail: hangauer@buffalo.edu
[**] G.K. is grateful to the EU for providing an ERC Advanced Grant
(DrugProfilBind 268145). We thank the beamline support staff at
the Helmholtz-Zentrum Berlin, Bessy II for radiation time and their
advice during data collection. We thank the HZB for financial
support of travel costs.
Supporting information for this article (experimental procedures for
the crystallization and soaking of the ligands, data collection and
processing, crystal structure determination and refinement, kinetic
assay, isothermal titration calorimetry, and the synthesis of ligands
1–8) is available on the WWW under http://dx.doi.org/10.1002/
anie.201208561.
1822
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828

Angewandte
Chemie
preferentially hydrophobic residues (e.g. the side chains of
Val, Leu, or Phe), and determines substrate specificity.[29] The
neighboring, more shallow and bowl-shaped S2’ pocket is also
of hydrophobic nature; however, it is more easily accessible to
water molecules from the bulk phase. Nevertheless, with
respect to the binding of ligand side chains, S2’ can host groups
very similar to those preferentially hosted by the S1’ pocket.
Interestingly, across ligand series with P1’ or P2’ side chains
modified from Gly to Leu the inhibitory potency increases
strongly by 800-fold for the S1’ pocket whereas for S2’ only
a 50-fold increase is experienced.[6] Furthermore, increasing
the hydrophobic interactions in the S1’ pocket is strikingly
enthalpy-driven and not, as assumed for the classical hydro-
phobic effect, entropically beneficial.[6] The favorable
enthalpic signal, observed for growing hydrophobicity of the
P1’ substituent in the congeneric series, was attributed to
“poor solvation” of the S1’ pocket. The latter property most
likely does not relate to a water-free enzyme pocket in the
unbound state but to the accommodation of several highly
mobile water molecules which are scattered over multiple
positions and thus are hardly detectable by crystal structure
analysis. The situation is very different for the S2’ pocket
where, based on recently determined high-resolution crystal
structures, a sophisticated arrangement of a complex water
network is observed next to the S2’ pocket that exerts
a dominant influence on the binding properties of the
accommodated ligands.[30,31]
provide insights into the driving forces associated with
hydrophobic binding to the S2’ pocket of TLN.
The crystal structures of eight TLN inhibitors containing
the Cbz-Gly-(PO2)À-l-Leu-l-X scaffold (Cbz = carboxyben-
zyl, X = Gly 1, Ala 2, Et-Gly 3, Val 4, nPr-Gly 5, Ile 6, Leu 7,
Phe 8; see Scheme 1) in complex with thermolysin have been
determined at high resolution (1.28–1.66 ꢁ). Crystal struc-
tures with ligands 1, 2, and 7 were studied previously.[6,30] For
TLN-7, the original data have been newly refined so that the
same protocol could be applied to all complexes. As the
binding mode of the parent scaffold has been already
described,[30] we will briefly discuss the predominant inter-
actions of the ligand to TLN and focus only on novel
structural features.
The electron density of the scaffold is well defined for all
the studied inhibitors (see 3 in Figure S1 in the Supporting
Information as a representative of the series). No significant
changes in the binding mode are observed among all ligands
(Figure S2). The Cbz moiety binds to the unspecific S1 pocket
and the central phosphonamidate group coordinates in
monodentate fashion through one of its oxygen atoms to
the zinc ion (2.0 ꢁ), whereas the other is hydrogen-bonded to
OE1 of Glu143. The leucyl P1’ and the structurally varied P2’
ligand side chains interact with the hydrophobic environment
of the S1’ and S2’ pockets, respectively (for further details see
the Supporting Information).
Major differences among the complexes are evident in the
water network adjacent to the S2’ pocket. This network is
perturbed and modulated by the size of the P2’ substituent.
Unfortunately, not all complexes could be determined at the
same resolution; the complex of the valyl derivative 4 even
shows some disorder in this crucial region. For the related n-
propylglycyl 5, the residual difference density indicates that
some disorder of the side chain might occur. We however
decided to describe the final density with one model. Local
disorder makes it difficult to reliably detect water molecules
with increasing distance from the protein surface, or from
polar ligand functional groups, particularly if chains of
contiguously connected water molecules are analyzed. There-
fore, the diffraction properties of the crucial water molecules
have been thoroughly inspected by difference electron
density maps (FoÀFc) to examine the accuracy and reliability
of the hydration properties of the S2’ pocket. Particularly, the
B-factors and occupancies which are highly correlated have
been regarded with care. These limitations complicate
a straight-forward comparison of the absolute numbers of
water molecules across the ligand series, especially consider-
ing the relative inventory of released or picked-up water
molecules. The distances along the water network vary and
may even correlate with the strength of formed hydrogen
bonds. However, the determined spatial accuracy of individ-
ual water positions can be affected by, for example, residual
mobility, disorder, and partial occupancy which limits posi-
tional accuracy. Therefore, we refrained from any detailed
analysis of the variation in H-bond length.
Homans et al.[7,8] reported a similar enthalpy-driven
hydrophobic interaction which has also been attributed to
suboptimal solvation of a protein pocket. The desolvation
enthalpy of the protein binding pocket is greatly decreased so
that overall the thermodynamic signature shifts to an over-
whelmingly enthalpy-driven Gibbs free energy of binding.
Snyder et al.[9] also reported an enthalpy-driven thermody-
namic profile for heterocyclic aromatic sulfonamides with
increasing hydrophobic properties against carbonic anhy-
drase. They explain the observed differences with changes in
the number and organization of well-ordered water molecules
in the binding site. In a study regarding the displacement of
several well-ordered water molecules from the S3/4 pocket of
thrombin by increasingly hydrophobic P3 substituents of
peptidomimetic inhibitors, we observed an entropy-driven
signal.[10] In the field of host–guest chemistry, several exam-
ples of complex formation have been reported to be driven
either by enthalpy or entropy improvement.[11,12] These
studies indicate that the thermodynamic signature of hydro-
phobic binding is determined by changes in the water
structure, and will depend on the properties of the water
molecules being reorganized during the binding process.
Since the S1’ and S2’ pockets of TLN exhibit opposite
features with respect to the observed solvation patterns, but
can host chemically similar ligand side chains, we embarked
on the study of a congeneric series of peptidomimetics which
contain a step-by-step series of modifications in the P2’ side
chain. High-resolution crystal structures of the protein–ligand
complexes were analyzed in order to investigate the changes
in the water structure, as it is modulated by the interactions of
the P2’ side chains. Isothermal titration calorimetry (ITC) data
were recorded to complement crystallographic findings and to
Virtually the same solvation pattern is observed for all
complexes next to the Cbz carbonyl and the negatively
charged terminal carboxylate group (Figure S2 right, upper
part). A network of at least seven mutually connected water
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
1823

.
Angewandte
Communications
molecules mediates interactions between the latter two ligand
functional groups and Asp226, Asn227, and Asn112. In two
complexes water molecules corresponding to the second
solvation shell are indicated (TLN-4, TLN-5).
tions. This partial disorder is translated to the neighboring
Leu202 residue as its isobutyl side chain adopts two con-
formations which refine to 55% and 45% occupancy (Fig-
ure S4). Both conformations occur in a correlated manner due
to mutual steric interference. When one considers the water
molecules picked up by TLN–2 and –3 compared to TLN–1,
the complexes with 4 and 5 show the water molecule capping
the position of the carboxylate group (Figure 1d,e, circled in
green).
The crystal structure of the complex with the Ile
derivative 6 (Figure 1 f) suggests again a contiguously con-
nected water network which wraps around the terminal
hydrophobic group as is also seen in TLN–2 and –3. The
network takes a more extended detour around the butyl
group than in TLN–2. TLN–7, reported in a previous study
(PDB code 3FWD[6]), was re-refined in the present study in
order to apply exactly the same refinement protocol and
program suite. Even though the indicated network is not as
complete as that for TLN–6, a related pattern is indicated for
TLN–7 (Figure 1 g). For both complexes the water molecule
at the position capping the carboxylate group is no longer
observed, in contrast to the complexes with 2, 3, 4, and 5
(Figure 1b–e, circled in green). In the latter crystal structures
the shortest distance between the capping water and the alkyl
side chain amounts to 3.80–3.85 ꢁ, in TLN–4 with the
branched valyl group, present with disordered geometry, the
water is slightly shifted (Figure S4a). If one takes the mean
A more complex pattern is observed next to the area of
the S2’ pocket where the growing P2’ side chain extends and
perturbs the water network (Figure S2, right, lower part). In
a previous study the crystal structures of TLN-1 and TLN-2
were compared.[30] The glycine derivative 1 shows two water
molecules hydrogen-bonded to the backbone carbonyl group
of Asn111 which are displaced from the TLN-2 complex
(Figure 1a, circled in cyan) due to steric conflicts with the
attached methyl group in the alanyl derivative 2. In contrast,
the latter recruits two additional water molecules (Figure 1b,
circled in yellow and green) which are picked up and form
favorable van der Waals contacts with the terminal methyl
group. In TLN–2 the water network establishes a contiguously
connected water chain from the ligandꢀs carboxylate group to
=
Asn111(C O), whereas in TLN–1 the water network is
disrupted (Figure 1a,b).
A comparison of the water networks in TLN–2 and TLN–
3 (Figure 1b,c) suggests nearly identical hydration patterns,
whereas those in TLN–4 and TLN–5 (Figure 1d,e) seem to
deviate and are disconnected at the lower left rim of the
pocket. Nonetheless, the valyl 4 and n-propyl 5 derivatives
display very similar water network patterns. In TLN–4, the
isopropyl side chain is scattered over at least two conforma-
Figure 1. Binding modes of the ligands 1–8. Each complex is shown with a different color, heteroatoms in atom-type color coding, water
molecules as spheres with the same color as the parent structure. In TLN–1 two water molecules (circled in cyan) are present that are replaced in
the other complexes as a result of the steric requirement of the growing P2’ substituent. TLN–1 shows a break in the contiguously connected
water network (red arrow) which is closed in TLN–2 and TLN–3 by the pick-up of an additional water molecule (circled in yellow) and which is
stabilized by both the favorable van der Waals contacts with the P2’ methyl or ethyl group in 2 or 3, and the H-bonds with other water molecules.
Similar favorable van der Waals interactions help accommodate a water molecule at a position capping the ligand’s carboxylate group (circled in
green) in complexes with 2, 3, 4, and 5. In the complexes with 6, 7, and 8 this water molecule is repelled, whereas 8 picks up a water molecule
(circled in magenta) next to the benzyl moiety of the ligand’s P2’ substituent.
1824
www.angewandte.org
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828

Angewandte
Chemie
position of the capping water in these complexes to calculate
putative distances to the more bulky butyl side chains in
TLN–6 and –7, an 0.3 ꢁ shorter contact would be created.
Supposedly, this contact distance is too short and sterically
unfavorable so that the capping water is no longer found in
TLN–6 and –7.
Changes in the thermodynamic parameters for binding
across the inhibitor series were determined by ITC. Absolute
thermodynamic values of the inhibitor binding could not be
determined because ligand binding is superimposed by the
displacement of the cleavage product Val-Lys resulting from
autoprotolysis at high TLN concentrations. In this respect, all
the measured thermodynamic values include a constant
contribution resulting from the displacement of Val-Lys.
This constant contribution of the dipeptide displacement is
cancelled out in the relative values. Enzyme kinetic inhibition
data of all inhibitors 1–8 were obtained, and demonstrated
binding free energy changes similar in magnitude to the
changes recorded by ITC, despite a constant offset (see
Table S3 in the Supporting Information). Kinetic inhibition
data, therefore, confirm the data obtained in ITC experi-
ments.
Furthermore, a buffer dependence of our recorded ITC
data was measured which shows that all complexes pick up
one proton per formed protein–ligand complex. The data
indicate that Glu143 next to the catalytic zinc ion changes its
protonation state upon inhibitor binding (see the Supporting
Information). As this residue is not directly involved in the
binding of the P2’ portion, the thermodynamic data will be
affected for all complexes in the same way, thus in a relative
comparison also this contribution cancels out.
Finally, TLN–8 with a terminal benzyl moiety shows the
least amount of ordered water molecules next to the S2’
pocket. It seems that nearly all water molecules observed in
other complexes close to the lower rim of the S2’ pocket are
either repelled or not sufficiently well ordered. Most likely
this correlates with increasing steric requirements of the
benzyl group which fills the S2’ pocket quite significantly.
Remarkably, however, one water molecule, which is also
found in TLN–2 and –3 (Figure 1h, circled in magenta) and
occupies a site very close to the ligand, can be observed in the
TLN–8 complex. This site is clearly not accommodated in
TLN–6 and –7, whereas in TLN–8 a water molecule is found
at this site stabilizing interactions (3.2 ꢁ) with the p-system of
the ligandꢀs neighboring phenyl ring (Figure S5). Interest-
ingly, the water molecule at the position capping the ligandꢀs
carboxylate group is also missing in TLN–8, apparently
because of steric interference with the ligandꢀs terminal P2’
substituent.
The bulky benzyl group in TLN–8 has an impact on the
neighboring protein molecule. It interferes with the carbonyl
group of Asn111, which is pushed to a different position,
giving rise to a second conformation (Figure S5). This
perturbation is accompanied by a partial loss of the planarity
of the peptide bond between Asn111 and Asn112 which is not
observed in the other complexes of the series (w angle
deviates from planarity by 10.48 and À17.38; more details in
Table S2 in the Supporting Information). Apparently, the
carbonyl oxygen evades in two directions to create enough
space for the large benzyl side chain of 8.
Figure 2 illustrates that, apart from TLN–1, binding
becomes increasingly more entropic as the size of the
attached hydrophobic P2’ substituent increases.[32] Simultane-
ously, significant enthalpy–entropy compensation is observed.
As a consequence, across the series the net changes in the free
energy are much smaller than those observed for either
enthalpy or entropy.
A remarkable relative gain in the potency (DDG1/2
=
À5.7 kJmolÀ1) is obtained going from 1 to 2. This gain is
mainly due to an improvement in enthalpy (DDH1/2
=
Figure 2. The diagram shows the observed thermodynamic results for DG (blue), DH (green), and ÀTDS (red) as obtained by ITC. The
experiments were performed in HEPES buffer and not corrected for superimposed protonation steps and replacement of the autocleavage product
Val-Lys. Right: To show the relative differences, mutual enthalpy–entropy compensation leading to minor changes in free energy and stepwise
changes in terms of related pairs. The thermodynamic data are depicted in an alternative way and the numbers shown give the relative differences
between neighboring ligands in the diagram.
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
1825

.
Angewandte
Communications
À13.4 kJmolÀ1) and the effect is partly compensated by the
conserved water molecules and mediates a complex network
between ligand and protein functional groups, does not show
any significant changes across the series (Figure S2, right
upper part). In addition, no change can be detected for the
contacts to the glycerol and DMSO molecules picked up from
the cryo buffer. Thus, the only differences occur next to the
hydrophobic P2’ substituents which increase in size (Figure S2,
right, lower part). Here, the network of the adjacent water
molecules is highly perturbed.
A huge change in the thermodynamic profile is experi-
enced when a methyl group is added to the glycine derivative
1 resulting in the alanine substituent in 2. One difference is
that the Gly derivative has in solution, prior to protein
binding, access to a larger conformational space than the Ala
analogue. Consequently, TLN–2 will experience a smaller loss
in entropy than TLN–1 upon binding. However, since the
thermodynamic profile of TLN–2 shows a larger, rather than
a smaller loss in entropy, additional effects are in operation. It
might be well possible that the conformational differences are
of minor entropic importance as ligands in a solvent cage will
hardly experience full flexibility. They could be as restricted in
their degrees of freedom as they would be at a binding site
which opens to the bulk solvent. As shown in our previous
study,[30] which also involves the non-carboxylated analogues
of 1 and 2, binding includes the rupture of the contiguously
connected water network which wraps around the terminal
methyl group in 2 (Figure 1a,b). It is apparent that the methyl
group provides favorable interaction sites for the two addi-
tional water molecules which are further stabilized in their
binding positions through van der Waals contacts. On the
other hand, two water molecules (Figure 1a, circled in cyan)
smaller entropic signal of 2 relative to 1 (-TDDS1/2
=
7.7 kJmolÀ1) (Table S5). Across the series, the Gibbs free
energy improves from 1 to 5, whereby 3, and 5 show the same
values, within experimental accuracy. The remaining ligands
6, 7, and 8 decrease slightly in affinity (Figure 2).
If one considers closely related ligand pairs, some system-
atic changes are observed. For example, individual ligands in
the pairs 2/3 (DDH2/3 = À1.2 kJmolÀ1, ÀTDDS2/3
À1.0 kJmolÀ1), 4/5 (DDH4/5 = À0.6 kJmolÀ1, ÀTDDS4/5
0.6 kJmolÀ1), and 6/7 (DDH6/7 = 0.9 kJmolÀ1, ÀTDDS6/7
=
=
=
À1.6 kJmolÀ1) exhibit very similar changes in their properties
relative to each other, whereas going from 3 to 4 (DDH3/4
=
2.6 kJmolÀ1, ÀTDDS3/4 = À2.5 kJmolÀ1), and from 5 to 6
(DDH5/6 = 7.7 kJmolÀ1
involves larger changes. Finally, the Phe derivative deviates
from the pair 6/7 and shows a thermodynamic signature with
balanced enthalpic and entropic portions.
and
ÀTDDS5/6 = À5.8 kJmolÀ1)
In the series of the peptidomimetic transition-state-
analogue inhibitors reported herein, the terminal hydro-
phobic substituent gradually penetrates the S2’ pocket of
thermolysin. This shallow bowl-shaped pocket is open to the
bulk solvent. It can host substituents up to the size of a benzyl
moiety. This group fills the pocket quite substantially. It even
pushes to the limit so that the backbone carbonyl group of the
adjacent peptide bond in the protein has to move out of the
position it occupies in the other complexes for steric reasons.
This carbonyl group evades in two directions and thereby
produces two alternative geometries which are likely to be
energetically disfavored.
Even though the hydrophobic surface increases continu-
ously in the series (1–8) from hydrogen in the Gly derivative
to phenyl in the Phe derivative by about 130 ꢁ2, the overall
Gibbs free energy improves only by À3.7 kJmolÀ1. This is
a minor contribution considering the rough estimate for the
free energy of dehydration of about À2 to À3 kJmolÀ1 per
methyl group that becomes buried upon protein binding.[33]
Purely based on surface patch considerations, we would
expect a much larger value as the hydrophobic effect for this
change. Interestingly, the affinity trend shows an optimum
with an ethyl (3), isopropyl (4), or n-propyl (5) substituent,
even though, in terms of size, these groups do not yet fill the
S2’ pocket completely. This indicates even more that simple
considerations based on hydrophobic surface patches buried
upon complex formation break down in the current analysis.
For the Phe derivative 8, the crystal structure indicates
disfavored conformations for Asn111 which will also influ-
ence the decrease in the affinity of this ligand. More
remarkable is the trend in enthalpy/entropy partitioning
(Figure 2) which is largest with the most potent inhibitors.
Changes are not consistently observed across the series, but
the structurally closely related pairs 2/3, 4/5, and 6/7 exhibit
very similar thermodynamic profiles (Figure 2). This suggests
for each pair similarities in the structural solvation patterns of
the individual complexes.
=
H-bonded in TLN-1 to Asn111(C O) are repelled from the
complex as a result of steric conflicts with the additional
methyl group in 2. The rupture of the contiguously connected
H-bonding network disfavors the exothermic binding of 1,
whereas the binding of 2 is entropically less favorable owing
to a stronger fixation of the water network. Accordingly,
going from 1 to the more hydrophobic 2 is enthalpy-driven
and could be classified—in formal terms—as a “nonclassical
hydrophobic effect”.
The thermodynamic signature of the ethyl derivative 3 is
nearly identical to that of 2. As both enthalpy and entropy
become more favorable relative to 2, DDG improves by
À2.2 kJmolÀ1, a value found in the typical range for favorably
placed additional methyl groups. It is in good agreement with
the estimated free energy of desolvation of a methyl group.
Again the additional degree of conformational freedom in 3
seems to be of minor importance. With respect to the water
network, TLN–3 is nearly identical with TLN–2. The isopro-
pyl and n-propyl derivatives 4 and 5 share again very similar
thermodynamic properties, but their enthalpy/entropy values
are much different from those of the previous pair 2/3. In
TLN–4 the branched and more rigid valyl ligand side chain is
distributed over two conformations and also a disorder of the
Leu202 side chain is detected. In TLN–5, the even more
flexible P2’ n-propyl group seems to be ordered; nonetheless
some disorder cannot be fully excluded by crystallography
though it is less evident. At the far end of the pocket two
water molecules mediating the water network in TLN–2 and
In the analysis of the binding modes of 1–8, the parent
scaffold remains virtually unchanged across the entire series.
Also the hydration pattern next to the ligandꢀs Cbz group and
the terminal carboxylate group, which involves at least seven
1826
www.angewandte.org
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828

Angewandte
Chemie
–3 cannot be detected and the network appears incomplete in
TLN–4 and –5. This should result, as seen for TLN–1, in an
enthalpic loss and an entropic gain. Both disorder and water
release support the entropic advantage of TLN–4 and –5;
however, with respect to the free energy, the latter advantage
is nearly canceled out. Nonetheless, it is remarkable that 4 and
5 not only share similar thermodynamic signatures, but also
the details of their local water networks show significant
resemblance. The hydrophobic effect related to the change
from 2/3 to 4/5 can be viewed as a “classical entropy-driven
hydrophobic effect”.
All four complexes TLN–2, –3, –4, and –5 host one water
molecule at a position capping the ligandꢀs carboxylate group.
They all show a nearly 10 kJmolÀ1 more enthalpic signal than
TLN–1, –6, --7, and –8 which lack this capping water. Its
spatial position should be energetically favored for electro-
static reasons. Frequent occupancy of such carboxylate–water
contacts was highlighted by Paulini et al.[34] in protein
structures and a clear preference for this geometry can be
found in the compilation of water–carboxylate contacts as
assembled in IsoStar.[35]
TLN–6 and –7 place a sec-butyl and isobutyl group,
respectively, into the S2’ pocket in an ordered fashion. Again
both complexes experience a very similar enthalpy/entropy
profile with strong enthalpic loss and entropic gain relative to
the pair 4/5. With respect to their water solvation pattern
quite similar networks are observed that wrap around the
terminal hydrophobic ligand side chain, and orient along the
rim of the S2’ pocket. Compared to TLN–2 and TLN–3, TLN–
6 has two water molecules shifted in position to more remote
sites in order to create an expanded network. The capping
water above the carboxylate group is no longer observed,
supposedly as a consequence of steric repulsion with the
larger butyl substituent. Apparently, the release of this special
water is one of the causes of the observed enthalpy loss and
entropy gain, which we would relate to the “classical hydro-
phobic effect”.
numbers of rotatable bonds also seem to be of minor
importance. The bound ligand side chains contribute together
with the protein to the newly formed complex surfaces. Water
molecules with their strong structure-determining properties
arrange along the new surfaces; they can even impact the
ligandꢀs bound geometry to find a best compromise with
respect to the formed water network. In total, all these
contributions starting from the separately solvated binding
partners to the formed complex describe quantitatively the
observed thermodynamic profiles. As thermolysin is a very
rigid protein, influences resulting from changes of the residual
mobility of protein residues or induced-fit adaptations, apart
from TLN–8, will be of minor importance. Even though our
discussion only intends to provide a qualitative correlation, it
is remarkable that complexes with side chains of comparable
size and number of rotatable bonds show very similar
thermodynamic signature.
The presented series can be used to correlate the details of
how the first solvation layer around a binding pocket impacts
ligand-binding affinity. It also shows that water networks can
have significant influence on modulating the structure–
activity relationships. Increasing the hydrophobicity of
ligand functional groups that bind in hydrophobic pockets is
usually discussed in terms of the hydrophobic effect. This
effect has been attributed to either an enthalpic or an entropic
signature.[6,7,11,12] Crystal structure analyses along with the
ITC data of the presented complex series underscore that
both, enthalpy and entropy, are involved in the hydrophobic
effect, and that many detailed structural phenomena deter-
mine the final overall signature. For example, if a contiguously
connected water network ruptures like in TLN–1, an
enthalpic loss and an entropic gain are experienced relative
to TLN–2.[26] The loss of the hydrogen bonds caused by this
rupture allows the system to activate and thus distribute
energy over more degrees of freedom. The displacement of
ordered water molecules can be entropically favorable. On
the other hand, the release of largely disordered waters can
also reveal a predominately enthalpic signal.[6–9] If parts of the
ligand are accommodated in pockets that are open to the bulk
solvent, and parts of the ligand are exposed to the water
phase, new binding sites for water molecules can be gener-
ated, for example, in our study the position capping the
carboxylate group or the site found on top of the phenyl ring.
The capping water seems to provide a significant contribu-
tion. The four complexes TLN–1, –6, –7, and –8 do not show
this water molecule supposedly because of steric interference
and they lose in the Gibbs free energy predominantly for
enthalpic reasons. This loss in enthalpy is partly compensated
by entropy as these complexes do not capture the water
molecule, a process which would be entropically unfavorable
if it were to occur. All these phenomena contribute on the
molecular level to the finally determined hydrophobic effect.
In summary, there are no universally valid reasons why the
hydrophobic effect should be predominantly “entropic” or
“enthalpic”; small structural changes in the binding features
of water molecules on the molecular level determine whether
hydrophobic binding is enthalpically or entropically driven.
Admittedly, this study reaches the limits of experimental
accuracy accomplishable in contemporary protein–ligand
Finally, the benzyl derivative 8 loses 2.9 kJmolÀ1 in DDG
with respect to 7. This price is paid in enthalpy, and correlates
with the steric clash of the phenyl group with the backbone
carbonyl group of Asn111. Furthermore, in this complex the
capping water is also repelled from the site above the
carboxylate group. The solvation structure at the far end of
the S2’ pocket appears to be thinned out; however, one water
site that is close to the ligand and has already been
accommodated in TLN–2 and –3 is newly populated (Fig-
ure S5a). This water molecule (Figure 1h, circled in magenta)
is most likely stabilized at this pivotal position perpendicular
to the terminal aromatic benzyl moiety by the interactions
with the phenyl group. Thus in this final case, compared to
TLN–6 and –7, the hydrophobic effect is paid in enthalpy and
even a new water binding site is populated.
The described structural features next to the S2’ pocket
show that simple models[36,37] based on buried ligands and
protein surface patches to describe beneficial desolvation
contributions arising from size differences of the ligands are
not sufficient to explain the observed thermodynamic signa-
ture across the series. Differences in the varying conforma-
tional properties of the P2’ side chains exhibiting different
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
1827

.
Angewandte
Communications
[9] P. W. Snyder, J. Mecinovic, D. T. Moustakas, S. W. Thomas, 3rd,
M. Harder, E. T. Mack, M. R. Lockett, A. Heroux, W. Sherman,
G. M. Whitesides, Proc. Natl. Acad. Sci. USA 2011, 108, 17889 –
17894.
[10] A. Biela, F. Sielaff, F. Terwesten, A. Heine, T. Steinmetzer, G.
Klebe, J. Med. Chem. 2012, 55, 6094 – 6110.
[11] D. B. Smithrud, T. B. Wyman, F. Diederich, J. Am. Chem. Soc.
1991, 113, 5420 – 5426.
[12] K. J. Breslauer, D. P. Remeta, W. Y. Chou, R. Ferrante, J. Curry,
D. Zaunczkowski, J. G. Snyder, L. A. Marky, Proc. Natl. Acad.
Sci. USA 1987, 84, 8922 – 8926.
[13] T. Young, R. Abel, B. Kim, B. J. Berne, R. A. Friesner, Proc.
Natl. Acad. Sci. USA 2007, 104, 808 – 813.
[14] J. M. Word, M. T. Geballe, A. Nicholls, Abstr. Pap. Am. Chem.
Soc. 2010, 240 (Meeting Abstract 280-Comp).
[15] J. T. Bukrinsky, M. J. Bjerrum, A. Kadziola, Biochemistry 1998,
37, 16555 – 16564.
[16] H. M. Holden, D. E. Tronrud, A. F. Monzingo, L. H. Weaver,
B. W. Matthews, Biochemistry 1987, 26, 8542 – 8553.
[17] D. E. Tronrud, A. F. Monzingo, B. W. Matthews, Eur. J. Biochem.
1986, 157, 261 – 268.
[18] P. A. Bartlett, C. K. Marlowe, Biochemistry 1983, 22, 4618 – 4624.
[19] M. Selkti, A. Tomas, J. F. Gaucher, T. Prange, M. C. Fournie-
Zaluski, H. Chen, B. P. Roques, Acta Crystallogr. Sect. D 2003,
59, 1200 – 1205.
structural work. Assignment of water molecules to the
difference electron density is crucial and depends on both
resolution and the local ordering phenomena. In the present
series, eight crystal structures indicate similar and internally
consistent solvation patterns which give confidence about the
reliability necessary to underscore their relevance. Our
structural data originate from crystal environments of flash-
cooled crystals which should capture a frozen image of the
structure at ambient temperature.[38] The thermodynamic
data were recorded in solution. Nevertheless, a very consis-
tent picture emerges. Surprising pairwise systematic changes
in the thermodynamic data are experienced for complexes of
related ligands, and they are convincingly well reflected by the
structural properties. The present study unravels small but
important details. Computational methods simulate molec-
ular properties at the atomic level, and are usually determined
by the summation of many small details.[39] However, details
such as those observed here are usually not regarded by these
computational methods as relevant, simply because we are
not fully aware of their importance for protein–ligand bind-
ing, structure–activity relationships, and rational drug design
in general. The study presented here sheds light on the role
these details play in the binding process. Taking such details
into account will make computational simulations not neces-
sarily simpler, but hopefully better in predicting ligand-
binding affinity, as well as the other molecular recognition
aspects.
[20] J. F. Gaucher, M. Selkti, G. Tiraboschi, T. Prange, B. P. Roques,
A. Tomas, M. C. Fournie-Zaluski, Biochemistry 1999, 38, 12569 –
12576.
[21] T. Takeuchi, A. Bottcher, C. M. Quezada, T. J. Meade, H. B.
Gray, Bioorg. Med. Chem. 1999, 7, 815 – 819.
[22] K. J. Lee, D. H. Kim, Bioorg. Med. Chem. Lett. 1998, 8, 323 – 326.
[23] D. G. Hangauer, A. F. Monzingo, B. W. Matthews, Biochemistry
1984, 23, 5730 – 5741.
[24] A. C. English, C. R. Groom, R. E. Hubbard, Protein Eng. 2001,
14, 47 – 59.
[25] J. Behnen, H. Koster, G. Neudert, T. Craan, A. Heine, G. Klebe,
ChemMedChem 2012, 7, 248 – 261.
[26] L. Englert, K. Silber, H. Steuber, S. Brass, B. Over, H. D. Gerber,
A. Heine, W. E. Diederich, G. Klebe, ChemMedChem 2010, 5,
930 – 940.
[27] B. W. Matthews, L. H. Weaver, W. R. Kester, J. Biol. Chem.
1974, 249, 8030 – 8044.
Experimental Section
Coordinates and structure factors have been deposited in the Protein
Data Bank with the following accession codes: TLN–3 complex 3T87;
TLN–4 complex 3T8H; TLN–5 complex 3T8C; TLN–6 complex
3T8D; and TLN–8 complex 4D9W. TLN–1 and TLN–2 were
previously reported as 3T8G and 3T74, respectively. TLN–7 (4H57)
has been newly refined and deposited replacing the former refine-
ment results.
[28] M. A. Holmes, B. W. Matthews, J. Mol. Biol. 1982, 160, 623 – 639.
[29] K. Morihara, Biochem. Biophys. Res. Commun. 1967, 26, 656 –
661.
Received: October 24, 2012
Published online: January 2, 2013
[30] A. Biela, M. Betz, A. Heine, G. Klebe, ChemMedChem 2012, 7,
1423 – 1434.
[31] N. N. Nasief, H. Tan, J. Kong, D. G. Hangauer, J. Med. Chem.
2012, 55, 8283 – 8302.
[32] N. N. Nasief, A. Biela, G. Klebe, D. Hangauer in 242nd ACS
National Meeting & Exposition, Denver, CO, United States,
2011.
[33] C. S. Leung, S. S. Leung, J. Tirado-Rives, W. L. Jorgensen, J.
Med. Chem. 2012, 55, 4489 – 4500.
[34] R. Paulini, K. Mꢂller, F. Diederich, Angew. Chem. 2005, 117,
1820 – 1839; Angew. Chem. Int. Ed. 2005, 44, 1788 – 1805.
[35] I. J. Bruno, J. C. Cole, J. P. Lommerse, R. S. Rowland, R. Taylor,
M. L. Verdonk, J. Comput.-Aided Mol. Des. 1997, 11, 525 – 537.
[36] M. H. Abraham, G. S. Whiting, R. Fuchs, E. J. Chambers, J.
Chem. Soc. Perkin Trans. 2 1990, 291 – 300.
[37] W. L. Jorgensen, J. P. Ulmschneider, J. Tirado-Rives, J. Phys.
Chem. B 2004, 108, 16264 – 16270.
[38] M. Nakasako, Philos. Trans. R. Soc. London Ser. B 2004, 359,
1191 – 1204.
Keywords: crystal structure analysis · enthalpy–
entropy compensation · hydrophobic effect · protein–
ligand interactions · water solvation
.
[1] “hydrophobic interaction:” M. Nic, J. Jirat, B. Kosata, IUPAC
Compendium of Chemical Terminology (Online ed.), 2006, DOI:
10.1351/goldbook.H02907.
[2] D. Chandler, Nature 2005, 437, 640 – 647.
[3] W. Kauzmann, Adv. Protein Chem. 1959, 14, 1 – 63.
[4] M. Charton, B. I. Charton, J. Theor. Biol. 1982, 99, 629 – 644.
[5] G. M. Whitesides, V. M. Krishnamurthy, Q. Rev. Biophys. 2005,
38, 385 – 395.
[6] L. Englert, A. Biela, M. Zayed, A. Heine, D. Hangauer, G.
Klebe, Biochim. Biophys. Acta Gen. Subj. 2010, 1800, 1192 –
1202.
[7] R. J. Bingham, J. B. Findlay, S. Y. Hsieh, A. P. Kalverda, A.
Kjellberg, C. Perazzolo, S. E. Phillips, K. Seshadri, C. H. Trinh,
W. B. Turnbull, G. Bodenhausen, S. W. Homans, J. Am. Chem.
Soc. 2004, 126, 1675 – 1681.
[39] D. J. Huggins, W. Sherman, B. Tidor, J. Med. Chem. 2012, 55,
1424 – 1444.
[8] S. W. Homans, Drug Discovery Today 2007, 12, 534 – 539.
1828
www.angewandte.org
ꢀ 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2013, 52, 1822 –1828